config tweaks for batching

This reverts commit fa098c551a.

bits.go

@@ -5,6 +5,7 @@ import (
 	"github.com/sirupsen/logrus"
 )
 
+// TODO: Pretty sure this is just all sorts of racy now, we need it to be atomic
 type Bits struct {
 	length             uint64
 	current            uint64
@@ -43,7 +44,7 @@ func (b *Bits) Check(l logrus.FieldLogger, i uint64) bool {
 	}
 
 	// Not within the window
-	l.Debugf("rejected a packet (top) %d %d\n", b.current, i)
+	l.Error("rejected a packet (top) %d %d\n", b.current, i)
 	return false
 }
 

cert/cert.go

@@ -58,6 +58,9 @@ type Certificate interface {
 	// PublicKey is the raw bytes to be used in asymmetric cryptographic operations.
 	PublicKey() []byte
 
+	// MarshalPublicKeyPEM is the value of PublicKey marshalled to PEM
+	MarshalPublicKeyPEM() []byte
+
 	// Curve identifies which curve was used for the PublicKey and Signature.
 	Curve() Curve
 

cert/cert_v1.go

@@ -83,6 +83,10 @@ func (c *certificateV1) PublicKey() []byte {
 	return c.details.publicKey
 }
 
+func (c *certificateV1) MarshalPublicKeyPEM() []byte {
+	return marshalCertPublicKeyToPEM(c)
+}
+
 func (c *certificateV1) Signature() []byte {
 	return c.signature
 }

cert/cert_v1_test.go

@@ -1,6 +1,7 @@
 package cert
 
 import (
+	"crypto/ed25519"
 	"fmt"
 	"net/netip"
 	"testing"
@@ -13,6 +14,7 @@ import (
 )
 
 func TestCertificateV1_Marshal(t *testing.T) {
+	t.Parallel()
 	before := time.Now().Add(time.Second * -60).Round(time.Second)
 	after := time.Now().Add(time.Second * 60).Round(time.Second)
 	pubKey := []byte("1234567890abcedfghij1234567890ab")
@@ -60,6 +62,58 @@ func TestCertificateV1_Marshal(t *testing.T) {
 	assert.Equal(t, nc.Groups(), nc2.Groups())
 }
 
+func TestCertificateV1_PublicKeyPem(t *testing.T) {
+	t.Parallel()
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := ed25519.PublicKey("1234567890abcedfghij1234567890ab")
+
+	nc := certificateV1{
+		details: detailsV1{
+			name:           "testing",
+			networks:       []netip.Prefix{},
+			unsafeNetworks: []netip.Prefix{},
+			groups:         []string{"test-group1", "test-group2", "test-group3"},
+			notBefore:      before,
+			notAfter:       after,
+			publicKey:      pubKey,
+			isCA:           false,
+			issuer:         "1234567890abcedfghij1234567890ab",
+		},
+		signature: []byte("1234567890abcedfghij1234567890ab"),
+	}
+
+	assert.Equal(t, Version1, nc.Version())
+	assert.Equal(t, Curve_CURVE25519, nc.Curve())
+	pubPem := "-----BEGIN NEBULA X25519 PUBLIC KEY-----\nMTIzNDU2Nzg5MGFiY2VkZmdoaWoxMjM0NTY3ODkwYWI=\n-----END NEBULA X25519 PUBLIC KEY-----\n"
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), pubPem)
+	assert.False(t, nc.IsCA())
+
+	nc.details.isCA = true
+	assert.Equal(t, Curve_CURVE25519, nc.Curve())
+	pubPem = "-----BEGIN NEBULA ED25519 PUBLIC KEY-----\nMTIzNDU2Nzg5MGFiY2VkZmdoaWoxMjM0NTY3ODkwYWI=\n-----END NEBULA ED25519 PUBLIC KEY-----\n"
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), pubPem)
+	assert.True(t, nc.IsCA())
+
+	pubP256KeyPem := []byte(`-----BEGIN NEBULA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA P256 PUBLIC KEY-----
+`)
+	pubP256Key, _, _, err := UnmarshalPublicKeyFromPEM(pubP256KeyPem)
+	require.NoError(t, err)
+	nc.details.curve = Curve_P256
+	nc.details.publicKey = pubP256Key
+	assert.Equal(t, Curve_P256, nc.Curve())
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.True(t, nc.IsCA())
+
+	nc.details.isCA = false
+	assert.Equal(t, Curve_P256, nc.Curve())
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.False(t, nc.IsCA())
+}
+
 func TestCertificateV1_Expired(t *testing.T) {
 	nc := certificateV1{
 		details: detailsV1{

cert/cert_v2.go

@@ -114,6 +114,10 @@ func (c *certificateV2) PublicKey() []byte {
 	return c.publicKey
 }
 
+func (c *certificateV2) MarshalPublicKeyPEM() []byte {
+	return marshalCertPublicKeyToPEM(c)
+}
+
 func (c *certificateV2) Signature() []byte {
 	return c.signature
 }

cert/cert_v2_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestCertificateV2_Marshal(t *testing.T) {
+	t.Parallel()
 	before := time.Now().Add(time.Second * -60).Round(time.Second)
 	after := time.Now().Add(time.Second * 60).Round(time.Second)
 	pubKey := []byte("1234567890abcedfghij1234567890ab")
@@ -75,6 +76,58 @@ func TestCertificateV2_Marshal(t *testing.T) {
 	assert.Equal(t, nc.Groups(), nc2.Groups())
 }
 
+func TestCertificateV2_PublicKeyPem(t *testing.T) {
+	t.Parallel()
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := ed25519.PublicKey("1234567890abcedfghij1234567890ab")
+
+	nc := certificateV2{
+		details: detailsV2{
+			name:           "testing",
+			networks:       []netip.Prefix{},
+			unsafeNetworks: []netip.Prefix{},
+			groups:         []string{"test-group1", "test-group2", "test-group3"},
+			notBefore:      before,
+			notAfter:       after,
+			isCA:           false,
+			issuer:         "1234567890abcedfghij1234567890ab",
+		},
+		publicKey: pubKey,
+		signature: []byte("1234567890abcedfghij1234567890ab"),
+	}
+
+	assert.Equal(t, Version2, nc.Version())
+	assert.Equal(t, Curve_CURVE25519, nc.Curve())
+	pubPem := "-----BEGIN NEBULA X25519 PUBLIC KEY-----\nMTIzNDU2Nzg5MGFiY2VkZmdoaWoxMjM0NTY3ODkwYWI=\n-----END NEBULA X25519 PUBLIC KEY-----\n"
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), pubPem)
+	assert.False(t, nc.IsCA())
+
+	nc.details.isCA = true
+	assert.Equal(t, Curve_CURVE25519, nc.Curve())
+	pubPem = "-----BEGIN NEBULA ED25519 PUBLIC KEY-----\nMTIzNDU2Nzg5MGFiY2VkZmdoaWoxMjM0NTY3ODkwYWI=\n-----END NEBULA ED25519 PUBLIC KEY-----\n"
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), pubPem)
+	assert.True(t, nc.IsCA())
+
+	pubP256KeyPem := []byte(`-----BEGIN NEBULA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA P256 PUBLIC KEY-----
+`)
+	pubP256Key, _, _, err := UnmarshalPublicKeyFromPEM(pubP256KeyPem)
+	require.NoError(t, err)
+	nc.curve = Curve_P256
+	nc.publicKey = pubP256Key
+	assert.Equal(t, Curve_P256, nc.Curve())
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.True(t, nc.IsCA())
+
+	nc.details.isCA = false
+	assert.Equal(t, Curve_P256, nc.Curve())
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.False(t, nc.IsCA())
+}
+
 func TestCertificateV2_Expired(t *testing.T) {
 	nc := certificateV2{
 		details: detailsV2{

cert/pem.go

@@ -1,25 +1,34 @@
 package cert
 
 import (
+	"encoding/hex"
 	"encoding/pem"
 	"fmt"
+	"time"
 
 	"golang.org/x/crypto/ed25519"
 )
 
-const (
-	CertificateBanner                = "NEBULA CERTIFICATE"
-	CertificateV2Banner              = "NEBULA CERTIFICATE V2"
-	X25519PrivateKeyBanner           = "NEBULA X25519 PRIVATE KEY"
-	X25519PublicKeyBanner            = "NEBULA X25519 PUBLIC KEY"
-	EncryptedEd25519PrivateKeyBanner = "NEBULA ED25519 ENCRYPTED PRIVATE KEY"
-	Ed25519PrivateKeyBanner          = "NEBULA ED25519 PRIVATE KEY"
-	Ed25519PublicKeyBanner           = "NEBULA ED25519 PUBLIC KEY"
+const ( //cert banners
+	CertificateBanner   = "NEBULA CERTIFICATE"
+	CertificateV2Banner = "NEBULA CERTIFICATE V2"
+)
 
-	P256PrivateKeyBanner               = "NEBULA P256 PRIVATE KEY"
-	P256PublicKeyBanner                = "NEBULA P256 PUBLIC KEY"
+const ( //key-agreement-key banners
+	X25519PrivateKeyBanner = "NEBULA X25519 PRIVATE KEY"
+	X25519PublicKeyBanner  = "NEBULA X25519 PUBLIC KEY"
+	P256PrivateKeyBanner   = "NEBULA P256 PRIVATE KEY"
+	P256PublicKeyBanner    = "NEBULA P256 PUBLIC KEY"
+)
+
+/* including "ECDSA" in the P256 banners is a clue that these keys should be used only for signing */
+const ( //signing key banners
 	EncryptedECDSAP256PrivateKeyBanner = "NEBULA ECDSA P256 ENCRYPTED PRIVATE KEY"
 	ECDSAP256PrivateKeyBanner          = "NEBULA ECDSA P256 PRIVATE KEY"
+	ECDSAP256PublicKeyBanner           = "NEBULA ECDSA P256 PUBLIC KEY"
+	EncryptedEd25519PrivateKeyBanner   = "NEBULA ED25519 ENCRYPTED PRIVATE KEY"
+	Ed25519PrivateKeyBanner            = "NEBULA ED25519 PRIVATE KEY"
+	Ed25519PublicKeyBanner             = "NEBULA ED25519 PUBLIC KEY"
 )
 
 // UnmarshalCertificateFromPEM will try to unmarshal the first pem block in a byte array, returning any non consumed
@@ -51,6 +60,16 @@ func UnmarshalCertificateFromPEM(b []byte) (Certificate, []byte, error) {
 
 }
 
+func marshalCertPublicKeyToPEM(c Certificate) []byte {
+	if c.IsCA() {
+		return MarshalSigningPublicKeyToPEM(c.Curve(), c.PublicKey())
+	} else {
+		return MarshalPublicKeyToPEM(c.Curve(), c.PublicKey())
+	}
+}
+
+// MarshalPublicKeyToPEM returns a PEM representation of a public key used for ECDH.
+// if your public key came from a certificate, prefer Certificate.PublicKeyPEM() if possible, to avoid mistakes!
 func MarshalPublicKeyToPEM(curve Curve, b []byte) []byte {
 	switch curve {
 	case Curve_CURVE25519:
@@ -62,6 +81,19 @@ func MarshalPublicKeyToPEM(curve Curve, b []byte) []byte {
 	}
 }
 
+// MarshalSigningPublicKeyToPEM returns a PEM representation of a public key used for signing.
+// if your public key came from a certificate, prefer Certificate.PublicKeyPEM() if possible, to avoid mistakes!
+func MarshalSigningPublicKeyToPEM(curve Curve, b []byte) []byte {
+	switch curve {
+	case Curve_CURVE25519:
+		return pem.EncodeToMemory(&pem.Block{Type: Ed25519PublicKeyBanner, Bytes: b})
+	case Curve_P256:
+		return pem.EncodeToMemory(&pem.Block{Type: P256PublicKeyBanner, Bytes: b})
+	default:
+		return nil
+	}
+}
+
 func UnmarshalPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
 	k, r := pem.Decode(b)
 	if k == nil {
@@ -73,7 +105,7 @@ func UnmarshalPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
 	case X25519PublicKeyBanner, Ed25519PublicKeyBanner:
 		expectedLen = 32
 		curve = Curve_CURVE25519
-	case P256PublicKeyBanner:
+	case P256PublicKeyBanner, ECDSAP256PublicKeyBanner:
 		// Uncompressed
 		expectedLen = 65
 		curve = Curve_P256
@@ -108,6 +140,101 @@ func MarshalSigningPrivateKeyToPEM(curve Curve, b []byte) []byte {
 	}
 }
 
+// Backward compatibility functions for older API
+func MarshalX25519PublicKey(b []byte) []byte {
+	return MarshalPublicKeyToPEM(Curve_CURVE25519, b)
+}
+
+func MarshalX25519PrivateKey(b []byte) []byte {
+	return MarshalPrivateKeyToPEM(Curve_CURVE25519, b)
+}
+
+func MarshalPublicKey(curve Curve, b []byte) []byte {
+	return MarshalPublicKeyToPEM(curve, b)
+}
+
+func MarshalPrivateKey(curve Curve, b []byte) []byte {
+	return MarshalPrivateKeyToPEM(curve, b)
+}
+
+// NebulaCertificate is a compatibility wrapper for the old API
+type NebulaCertificate struct {
+	Details   NebulaCertificateDetails
+	Signature []byte
+	cert      Certificate
+}
+
+// NebulaCertificateDetails is a compatibility wrapper for certificate details
+type NebulaCertificateDetails struct {
+	Name      string
+	NotBefore time.Time
+	NotAfter  time.Time
+	PublicKey []byte
+	IsCA      bool
+	Issuer    []byte
+	Curve     Curve
+}
+
+// UnmarshalNebulaCertificateFromPEM provides backward compatibility with the old API
+func UnmarshalNebulaCertificateFromPEM(b []byte) (*NebulaCertificate, []byte, error) {
+	c, rest, err := UnmarshalCertificateFromPEM(b)
+	if err != nil {
+		return nil, rest, err
+	}
+
+	issuerBytes, err := func() ([]byte, error) {
+		issuer := c.Issuer()
+		if issuer == "" {
+			return nil, nil
+		}
+		decoded, err := hex.DecodeString(issuer)
+		if err != nil {
+			return nil, fmt.Errorf("failed to decode issuer fingerprint: %w", err)
+		}
+		return decoded, nil
+	}()
+	if err != nil {
+		return nil, rest, err
+	}
+
+	pubKey := c.PublicKey()
+	if pubKey != nil {
+		pubKey = append([]byte(nil), pubKey...)
+	}
+
+	sig := c.Signature()
+	if sig != nil {
+		sig = append([]byte(nil), sig...)
+	}
+
+	return &NebulaCertificate{
+		Details: NebulaCertificateDetails{
+			Name:      c.Name(),
+			NotBefore: c.NotBefore(),
+			NotAfter:  c.NotAfter(),
+			PublicKey: pubKey,
+			IsCA:      c.IsCA(),
+			Issuer:    issuerBytes,
+			Curve:     c.Curve(),
+		},
+		Signature: sig,
+		cert:      c,
+	}, rest, nil
+}
+
+// IssuerString returns the issuer in hex format for compatibility
+func (n *NebulaCertificate) IssuerString() string {
+	if n.Details.Issuer == nil {
+		return ""
+	}
+	return hex.EncodeToString(n.Details.Issuer)
+}
+
+// Certificate returns the underlying certificate (read-only)
+func (n *NebulaCertificate) Certificate() Certificate {
+	return n.cert
+}
+
 // UnmarshalPrivateKeyFromPEM will try to unmarshal the first pem block in a byte array, returning any non
 // consumed data or an error on failure
 func UnmarshalPrivateKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {

cert/pem_test.go

@@ -177,6 +177,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 }
 
 func TestUnmarshalPublicKeyFromPEM(t *testing.T) {
+	t.Parallel()
 	pubKey := []byte(`# A good key
 -----BEGIN NEBULA ED25519 PUBLIC KEY-----
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
@@ -230,6 +231,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 }
 
 func TestUnmarshalX25519PublicKey(t *testing.T) {
+	t.Parallel()
 	pubKey := []byte(`# A good key
 -----BEGIN NEBULA X25519 PUBLIC KEY-----
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
@@ -240,6 +242,12 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 AAAAAAAAAAAAAAAAAAAAAAA=
 -----END NEBULA P256 PUBLIC KEY-----
+`)
+	oldPubP256Key := []byte(`# A good key
+-----BEGIN NEBULA ECDSA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA ECDSA P256 PUBLIC KEY-----
 `)
 	shortKey := []byte(`# A short key
 -----BEGIN NEBULA X25519 PUBLIC KEY-----
@@ -256,15 +264,22 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 -END NEBULA X25519 PUBLIC KEY-----`)
 
-	keyBundle := appendByteSlices(pubKey, pubP256Key, shortKey, invalidBanner, invalidPem)
+	keyBundle := appendByteSlices(pubKey, pubP256Key, oldPubP256Key, shortKey, invalidBanner, invalidPem)
 
 	// Success test case
 	k, rest, curve, err := UnmarshalPublicKeyFromPEM(keyBundle)
 	assert.Len(t, k, 32)
 	require.NoError(t, err)
-	assert.Equal(t, rest, appendByteSlices(pubP256Key, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, rest, appendByteSlices(pubP256Key, oldPubP256Key, shortKey, invalidBanner, invalidPem))
 	assert.Equal(t, Curve_CURVE25519, curve)
 
+	// Success test case
+	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
+	assert.Len(t, k, 65)
+	require.NoError(t, err)
+	assert.Equal(t, rest, appendByteSlices(oldPubP256Key, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, Curve_P256, curve)
+
 	// Success test case
 	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Len(t, k, 65)

cert_test/cert.go

@@ -114,33 +114,6 @@ func NewTestCert(v cert.Version, curve cert.Curve, ca cert.Certificate, key []by
 	return c, pub, cert.MarshalPrivateKeyToPEM(curve, priv), pem
 }
 
-func NewTestCertDifferentVersion(c cert.Certificate, v cert.Version, ca cert.Certificate, key []byte) (cert.Certificate, []byte) {
-	nc := &cert.TBSCertificate{
-		Version:        v,
-		Curve:          c.Curve(),
-		Name:           c.Name(),
-		Networks:       c.Networks(),
-		UnsafeNetworks: c.UnsafeNetworks(),
-		Groups:         c.Groups(),
-		NotBefore:      time.Unix(c.NotBefore().Unix(), 0),
-		NotAfter:       time.Unix(c.NotAfter().Unix(), 0),
-		PublicKey:      c.PublicKey(),
-		IsCA:           false,
-	}
-
-	c, err := nc.Sign(ca, ca.Curve(), key)
-	if err != nil {
-		panic(err)
-	}
-
-	pem, err := c.MarshalPEM()
-	if err != nil {
-		panic(err)
-	}
-
-	return c, pem
-}
-
 func X25519Keypair() ([]byte, []byte) {
 	privkey := make([]byte, 32)
 	if _, err := io.ReadFull(rand.Reader, privkey); err != nil {

cmd/nebula-service/main.go

@@ -65,8 +65,16 @@ func main() {
 	}
 
 	if !*configTest {
-		ctrl.Start()
-		ctrl.ShutdownBlock()
+		wait, err := ctrl.Start()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
+		wait()
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

cmd/nebula/main.go

@@ -3,6 +3,9 @@ package main
 import (
 	"flag"
 	"fmt"
+	"log"
+	"net/http"
+	_ "net/http/pprof"
 	"os"
 
 	"github.com/sirupsen/logrus"
@@ -58,10 +61,22 @@ func main() {
 		os.Exit(1)
 	}
 
+	go func() {
+		log.Println(http.ListenAndServe("0.0.0.0:6060", nil))
+	}()
+
 	if !*configTest {
-		ctrl.Start()
+		wait, err := ctrl.Start()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
 		notifyReady(l)
-		ctrl.ShutdownBlock()
+		wait()
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

connection_manager.go

@@ -354,6 +354,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 
 		if mainHostInfo {
 			decision = tryRehandshake
+
 		} else {
 			if cm.shouldSwapPrimary(hostinfo) {
 				decision = swapPrimary
@@ -460,10 +461,6 @@ func (cm *connectionManager) shouldSwapPrimary(current *HostInfo) bool {
 	}
 
 	crt := cm.intf.pki.getCertState().getCertificate(current.ConnectionState.myCert.Version())
-	if crt == nil {
-		//my cert was reloaded away. We should definitely swap from this tunnel
-		return true
-	}
 	// If this tunnel is using the latest certificate then we should swap it to primary for a bit and see if things
 	// settle down.
 	return bytes.Equal(current.ConnectionState.myCert.Signature(), crt.Signature())
@@ -478,34 +475,31 @@ func (cm *connectionManager) swapPrimary(current, primary *HostInfo) {
 	cm.hostMap.Unlock()
 }
 
-// isInvalidCertificate decides if we should destroy a tunnel.
-// returns true if pki.disconnect_invalid is true and the certificate is no longer valid.
-// Blocklisted certificates will skip the pki.disconnect_invalid check and return true.
+// isInvalidCertificate will check if we should destroy a tunnel if pki.disconnect_invalid is true and
+// the certificate is no longer valid. Block listed certificates will skip the pki.disconnect_invalid
+// check and return true.
 func (cm *connectionManager) isInvalidCertificate(now time.Time, hostinfo *HostInfo) bool {
 	remoteCert := hostinfo.GetCert()
 	if remoteCert == nil {
-		return false //don't tear down tunnels for handshakes in progress
+		return false
 	}
 
 	caPool := cm.intf.pki.GetCAPool()
 	err := caPool.VerifyCachedCertificate(now, remoteCert)
 	if err == nil {
-		return false //cert is still valid! yay!
-	} else if err == cert.ErrBlockListed { //avoiding errors.Is for speed
-		// Block listed certificates should always be disconnected
-		hostinfo.logger(cm.l).WithError(err).
-			WithField("fingerprint", remoteCert.Fingerprint).
-			Info("Remote certificate is blocked, tearing down the tunnel")
-		return true
-	} else if cm.intf.disconnectInvalid.Load() {
-		hostinfo.logger(cm.l).WithError(err).
-			WithField("fingerprint", remoteCert.Fingerprint).
-			Info("Remote certificate is no longer valid, tearing down the tunnel")
-		return true
-	} else {
-		//if we reach here, the cert is no longer valid, but we're configured to keep tunnels from now-invalid certs open
 		return false
 	}
+
+	if !cm.intf.disconnectInvalid.Load() && err != cert.ErrBlockListed {
+		// Block listed certificates should always be disconnected
+		return false
+	}
+
+	hostinfo.logger(cm.l).WithError(err).
+		WithField("fingerprint", remoteCert.Fingerprint).
+		Info("Remote certificate is no longer valid, tearing down the tunnel")
+
+	return true
 }
 
 func (cm *connectionManager) sendPunch(hostinfo *HostInfo) {
@@ -536,45 +530,15 @@ func (cm *connectionManager) sendPunch(hostinfo *HostInfo) {
 func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
 	cs := cm.intf.pki.getCertState()
 	curCrt := hostinfo.ConnectionState.myCert
-	curCrtVersion := curCrt.Version()
-	myCrt := cs.getCertificate(curCrtVersion)
-	if myCrt == nil {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("version", curCrtVersion).
-			WithField("reason", "local certificate removed").
-			Info("Re-handshaking with remote")
-		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
+	myCrt := cs.getCertificate(curCrt.Version())
+	if curCrt.Version() >= cs.initiatingVersion && bytes.Equal(curCrt.Signature(), myCrt.Signature()) == true {
+		// The current tunnel is using the latest certificate and version, no need to rehandshake.
 		return
 	}
-	peerCrt := hostinfo.ConnectionState.peerCert
-	if peerCrt != nil && curCrtVersion < peerCrt.Certificate.Version() {
-		// if our certificate version is less than theirs, and we have a matching version available, rehandshake?
-		if cs.getCertificate(peerCrt.Certificate.Version()) != nil {
-			cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
-				WithField("version", curCrtVersion).
-				WithField("peerVersion", peerCrt.Certificate.Version()).
-				WithField("reason", "local certificate version lower than peer, attempting to correct").
-				Info("Re-handshaking with remote")
-			cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], func(hh *HandshakeHostInfo) {
-				hh.initiatingVersionOverride = peerCrt.Certificate.Version()
-			})
-			return
-		}
-	}
-	if !bytes.Equal(curCrt.Signature(), myCrt.Signature()) {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("reason", "local certificate is not current").
-			Info("Re-handshaking with remote")
 
-		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
-		return
-	}
-	if curCrtVersion < cs.initiatingVersion {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("reason", "current cert version < pki.initiatingVersion").
-			Info("Re-handshaking with remote")
+	cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
+		WithField("reason", "local certificate is not current").
+		Info("Re-handshaking with remote")
 
-		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
-		return
-	}
+	cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
 }

connection_manager_test.go

@@ -446,6 +446,10 @@ func (d *dummyCert) PublicKey() []byte {
 	return d.publicKey
 }
 
+func (d *dummyCert) MarshalPublicKeyPEM() []byte {
+	return cert.MarshalPublicKeyToPEM(d.curve, d.publicKey)
+}
+
 func (d *dummyCert) Signature() []byte {
 	return d.signature
 }

connection_state.go

@@ -13,6 +13,8 @@ import (
 	"github.com/slackhq/nebula/noiseutil"
 )
 
+// TODO: In a 5Gbps test, 1024 is not sufficient. With a 1400 MTU this is about 1.4Gbps of window, assuming full packets.
+// 4092 should be sufficient for 5Gbps
 const ReplayWindow = 1024
 
 type ConnectionState struct {

control.go

@@ -2,9 +2,11 @@ package nebula
 
 import (
 	"context"
+	"errors"
 	"net/netip"
 	"os"
 	"os/signal"
+	"sync"
 	"syscall"
 
 	"github.com/sirupsen/logrus"
@@ -13,6 +15,16 @@ import (
 	"github.com/slackhq/nebula/overlay"
 )
 
+type RunState int
+
+const (
+	Stopped  RunState = 0 // The control has yet to be started
+	Started  RunState = 1 // The control has been started
+	Stopping RunState = 2 // The control is stopping
+)
+
+var ErrAlreadyStarted = errors.New("nebula is already started")
+
 // Every interaction here needs to take extra care to copy memory and not return or use arguments "as is" when touching
 // core. This means copying IP objects, slices, de-referencing pointers and taking the actual value, etc
 
@@ -26,6 +38,9 @@ type controlHostLister interface {
 }
 
 type Control struct {
+	stateLock sync.Mutex
+	state     RunState
+
 	f                      *Interface
 	l                      *logrus.Logger
 	ctx                    context.Context
@@ -49,10 +64,21 @@ type ControlHostInfo struct {
 	CurrentRelaysThroughMe []netip.Addr     `json:"currentRelaysThroughMe"`
 }
 
-// Start actually runs nebula, this is a nonblocking call. To block use Control.ShutdownBlock()
-func (c *Control) Start() {
+// Start actually runs nebula, this is a nonblocking call.
+// The returned function can be used to wait for nebula to fully stop.
+func (c *Control) Start() (func(), error) {
+	c.stateLock.Lock()
+	if c.state != Stopped {
+		c.stateLock.Unlock()
+		return nil, ErrAlreadyStarted
+	}
+
 	// Activate the interface
-	c.f.activate()
+	err := c.f.activate()
+	if err != nil {
+		c.stateLock.Unlock()
+		return nil, err
+	}
 
 	// Call all the delayed funcs that waited patiently for the interface to be created.
 	if c.sshStart != nil {
@@ -72,15 +98,33 @@ func (c *Control) Start() {
 	}
 
 	// Start reading packets.
-	c.f.run()
+	c.state = Started
+	c.stateLock.Unlock()
+	return c.f.run(c.ctx)
+}
+
+func (c *Control) State() RunState {
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+	return c.state
 }
 
 func (c *Control) Context() context.Context {
 	return c.ctx
 }
 
-// Stop signals nebula to shutdown and close all tunnels, returns after the shutdown is complete
+// Stop is a non-blocking call that signals nebula to close all tunnels and shut down
 func (c *Control) Stop() {
+	c.stateLock.Lock()
+	if c.state != Started {
+		c.stateLock.Unlock()
+		// We are stopping or stopped already
+		return
+	}
+
+	c.state = Stopping
+	c.stateLock.Unlock()
+
 	// Stop the handshakeManager (and other services), to prevent new tunnels from
 	// being created while we're shutting them all down.
 	c.cancel()
@@ -89,7 +133,7 @@ func (c *Control) Stop() {
 	if err := c.f.Close(); err != nil {
 		c.l.WithError(err).Error("Close interface failed")
 	}
-	c.l.Info("Goodbye")
+	c.state = Stopped
 }
 
 // ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled

e2e/helpers_test.go

@@ -29,6 +29,8 @@ type m = map[string]any
 
 // newSimpleServer creates a nebula instance with many assumptions
 func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
+	l := NewTestLogger()
+
 	var vpnNetworks []netip.Prefix
 	for _, sn := range strings.Split(sVpnNetworks, ",") {
 		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
@@ -54,25 +56,6 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 		budpIp[3] = 239
 		udpAddr = netip.AddrPortFrom(netip.AddrFrom16(budpIp), 4242)
 	}
-	return newSimpleServerWithUdp(v, caCrt, caKey, name, sVpnNetworks, udpAddr, overrides)
-}
-
-func newSimpleServerWithUdp(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, udpAddr netip.AddrPort, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
-	l := NewTestLogger()
-
-	var vpnNetworks []netip.Prefix
-	for _, sn := range strings.Split(sVpnNetworks, ",") {
-		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
-		if err != nil {
-			panic(err)
-		}
-		vpnNetworks = append(vpnNetworks, vpnIpNet)
-	}
-
-	if len(vpnNetworks) == 0 {
-		panic("no vpn networks")
-	}
-
 	_, _, myPrivKey, myPEM := cert_test.NewTestCert(v, cert.Curve_CURVE25519, caCrt, caKey, name, time.Now(), time.Now().Add(5*time.Minute), vpnNetworks, nil, []string{})
 
 	caB, err := caCrt.MarshalPEM()
@@ -146,109 +129,6 @@ func newSimpleServerWithUdp(v cert.Version, caCrt cert.Certificate, caKey []byte
 	return control, vpnNetworks, udpAddr, c
 }
 
-// newServer creates a nebula instance with fewer assumptions
-func newServer(caCrt []cert.Certificate, certs []cert.Certificate, key []byte, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
-	l := NewTestLogger()
-
-	vpnNetworks := certs[len(certs)-1].Networks()
-
-	var udpAddr netip.AddrPort
-	if vpnNetworks[0].Addr().Is4() {
-		budpIp := vpnNetworks[0].Addr().As4()
-		budpIp[1] -= 128
-		udpAddr = netip.AddrPortFrom(netip.AddrFrom4(budpIp), 4242)
-	} else {
-		budpIp := vpnNetworks[0].Addr().As16()
-		// beef for funsies
-		budpIp[2] = 190
-		budpIp[3] = 239
-		udpAddr = netip.AddrPortFrom(netip.AddrFrom16(budpIp), 4242)
-	}
-
-	caStr := ""
-	for _, ca := range caCrt {
-		x, err := ca.MarshalPEM()
-		if err != nil {
-			panic(err)
-		}
-		caStr += string(x)
-	}
-	certStr := ""
-	for _, c := range certs {
-		x, err := c.MarshalPEM()
-		if err != nil {
-			panic(err)
-		}
-		certStr += string(x)
-	}
-
-	mc := m{
-		"pki": m{
-			"ca":   caStr,
-			"cert": certStr,
-			"key":  string(key),
-		},
-		//"tun": m{"disabled": true},
-		"firewall": m{
-			"outbound": []m{{
-				"proto": "any",
-				"port":  "any",
-				"host":  "any",
-			}},
-			"inbound": []m{{
-				"proto": "any",
-				"port":  "any",
-				"host":  "any",
-			}},
-		},
-		//"handshakes": m{
-		//	"try_interval": "1s",
-		//},
-		"listen": m{
-			"host": udpAddr.Addr().String(),
-			"port": udpAddr.Port(),
-		},
-		"logging": m{
-			"timestamp_format": fmt.Sprintf("%v 15:04:05.000000", certs[0].Name()),
-			"level":            l.Level.String(),
-		},
-		"timers": m{
-			"pending_deletion_interval": 2,
-			"connection_alive_interval": 2,
-		},
-	}
-
-	if overrides != nil {
-		final := m{}
-		err := mergo.Merge(&final, overrides, mergo.WithAppendSlice)
-		if err != nil {
-			panic(err)
-		}
-		err = mergo.Merge(&final, mc, mergo.WithAppendSlice)
-		if err != nil {
-			panic(err)
-		}
-		mc = final
-	}
-
-	cb, err := yaml.Marshal(mc)
-	if err != nil {
-		panic(err)
-	}
-
-	c := config.NewC(l)
-	cStr := string(cb)
-	c.LoadString(cStr)
-
-	control, err := nebula.Main(c, false, "e2e-test", l, nil)
-
-	if err != nil {
-		panic(err)
-	}
-
-	return control, vpnNetworks, udpAddr, c
-}
-
 type doneCb func()
 
 func deadline(t *testing.T, seconds time.Duration) doneCb {

e2e/tunnels_test.go

@@ -4,16 +4,12 @@
 package e2e
 
 import (
-	"fmt"
-	"net/netip"
 	"testing"
 	"time"
 
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/cert_test"
 	"github.com/slackhq/nebula/e2e/router"
-	"github.com/stretchr/testify/assert"
-	"gopkg.in/yaml.v3"
 )
 
 func TestDropInactiveTunnels(t *testing.T) {
@@ -59,309 +55,3 @@ func TestDropInactiveTunnels(t *testing.T) {
 	myControl.Stop()
 	theirControl.Stop()
 }
-
-func TestCertUpgrade(t *testing.T) {
-	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
-	// under ideal conditions
-	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-	caB, err := ca.MarshalPEM()
-	if err != nil {
-		panic(err)
-	}
-	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-
-	ca2B, err := ca2.MarshalPEM()
-	if err != nil {
-		panic(err)
-	}
-	caStr := fmt.Sprintf("%s\n%s", caB, ca2B)
-
-	myCert, _, myPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
-	_, myCert2Pem := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
-
-	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
-	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
-
-	myControl, myVpnIpNet, myUdpAddr, myC := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert}, myPrivKey, m{})
-	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
-
-	// Share our underlay information
-	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
-	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
-
-	// Start the servers
-	myControl.Start()
-	theirControl.Start()
-
-	r := router.NewR(t, myControl, theirControl)
-	defer r.RenderFlow()
-
-	r.Log("Assert the tunnel between me and them works")
-	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-	r.Log("yay")
-	//todo ???
-	time.Sleep(1 * time.Second)
-	r.FlushAll()
-
-	mc := m{
-		"pki": m{
-			"ca":   caStr,
-			"cert": string(myCert2Pem),
-			"key":  string(myPrivKey),
-		},
-		//"tun": m{"disabled": true},
-		"firewall": myC.Settings["firewall"],
-		//"handshakes": m{
-		//	"try_interval": "1s",
-		//},
-		"listen":  myC.Settings["listen"],
-		"logging": myC.Settings["logging"],
-		"timers":  myC.Settings["timers"],
-	}
-
-	cb, err := yaml.Marshal(mc)
-	if err != nil {
-		panic(err)
-	}
-
-	r.Logf("reload new v2-only config")
-	err = myC.ReloadConfigString(string(cb))
-	assert.NoError(t, err)
-	r.Log("yay, spin until their sees it")
-	waitStart := time.Now()
-	for {
-		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
-		if c == nil {
-			r.Log("nil")
-		} else {
-			version := c.Cert.Version()
-			r.Logf("version %d", version)
-			if version == cert.Version2 {
-				break
-			}
-		}
-		since := time.Since(waitStart)
-		if since > time.Second*10 {
-			t.Fatal("Cert should be new by now")
-		}
-		time.Sleep(time.Second)
-	}
-
-	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
-
-	myControl.Stop()
-	theirControl.Stop()
-}
-
-func TestCertDowngrade(t *testing.T) {
-	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
-	// under ideal conditions
-	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-	caB, err := ca.MarshalPEM()
-	if err != nil {
-		panic(err)
-	}
-	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-
-	ca2B, err := ca2.MarshalPEM()
-	if err != nil {
-		panic(err)
-	}
-	caStr := fmt.Sprintf("%s\n%s", caB, ca2B)
-
-	myCert, _, myPrivKey, myCertPem := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
-	myCert2, _ := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
-
-	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
-	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
-
-	myControl, myVpnIpNet, myUdpAddr, myC := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert2}, myPrivKey, m{})
-	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
-
-	// Share our underlay information
-	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
-	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
-
-	// Start the servers
-	myControl.Start()
-	theirControl.Start()
-
-	r := router.NewR(t, myControl, theirControl)
-	defer r.RenderFlow()
-
-	r.Log("Assert the tunnel between me and them works")
-	//assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
-	//r.Log("yay")
-	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-	r.Log("yay")
-	//todo ???
-	time.Sleep(1 * time.Second)
-	r.FlushAll()
-
-	mc := m{
-		"pki": m{
-			"ca":   caStr,
-			"cert": string(myCertPem),
-			"key":  string(myPrivKey),
-		},
-		"firewall": myC.Settings["firewall"],
-		"listen":   myC.Settings["listen"],
-		"logging":  myC.Settings["logging"],
-		"timers":   myC.Settings["timers"],
-	}
-
-	cb, err := yaml.Marshal(mc)
-	if err != nil {
-		panic(err)
-	}
-
-	r.Logf("reload new v1-only config")
-	err = myC.ReloadConfigString(string(cb))
-	assert.NoError(t, err)
-	r.Log("yay, spin until their sees it")
-	waitStart := time.Now()
-	for {
-		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
-		c2 := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
-		if c == nil || c2 == nil {
-			r.Log("nil")
-		} else {
-			version := c.Cert.Version()
-			theirVersion := c2.Cert.Version()
-			r.Logf("version %d,%d", version, theirVersion)
-			if version == cert.Version1 {
-				break
-			}
-		}
-		since := time.Since(waitStart)
-		if since > time.Second*5 {
-			r.Log("wtf")
-		}
-		if since > time.Second*10 {
-			r.Log("wtf")
-			t.Fatal("Cert should be new by now")
-		}
-		time.Sleep(time.Second)
-	}
-
-	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
-
-	myControl.Stop()
-	theirControl.Stop()
-}
-
-func TestCertMismatchCorrection(t *testing.T) {
-	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
-	// under ideal conditions
-	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-
-	myCert, _, myPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
-	myCert2, _ := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
-
-	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
-	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
-
-	myControl, myVpnIpNet, myUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert2}, myPrivKey, m{})
-	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
-
-	// Share our underlay information
-	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
-	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
-
-	// Start the servers
-	myControl.Start()
-	theirControl.Start()
-
-	r := router.NewR(t, myControl, theirControl)
-	defer r.RenderFlow()
-
-	r.Log("Assert the tunnel between me and them works")
-	//assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
-	//r.Log("yay")
-	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-	r.Log("yay")
-	//todo ???
-	time.Sleep(1 * time.Second)
-	r.FlushAll()
-
-	waitStart := time.Now()
-	for {
-		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
-		c2 := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
-		if c == nil || c2 == nil {
-			r.Log("nil")
-		} else {
-			version := c.Cert.Version()
-			theirVersion := c2.Cert.Version()
-			r.Logf("version %d,%d", version, theirVersion)
-			if version == theirVersion {
-				break
-			}
-		}
-		since := time.Since(waitStart)
-		if since > time.Second*5 {
-			r.Log("wtf")
-		}
-		if since > time.Second*10 {
-			r.Log("wtf")
-			t.Fatal("Cert should be new by now")
-		}
-		time.Sleep(time.Second)
-	}
-
-	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
-
-	myControl.Stop()
-	theirControl.Stop()
-}
-
-func TestCrossStackRelaysWork(t *testing.T) {
-	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me     ", "10.128.0.1/24,fc00::1/64", m{"relay": m{"use_relays": true}})
-	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "relay  ", "10.128.0.128/24,fc00::128/64", m{"relay": m{"am_relay": true}})
-	theirUdp := netip.MustParseAddrPort("10.0.0.2:4242")
-	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServerWithUdp(cert.Version2, ca, caKey, "them   ", "fc00::2/64", theirUdp, m{"relay": m{"use_relays": true}})
-
-	//myVpnV4 := myVpnIpNet[0]
-	myVpnV6 := myVpnIpNet[1]
-	relayVpnV4 := relayVpnIpNet[0]
-	relayVpnV6 := relayVpnIpNet[1]
-	theirVpnV6 := theirVpnIpNet[0]
-
-	// Teach my how to get to the relay and that their can be reached via the relay
-	myControl.InjectLightHouseAddr(relayVpnV4.Addr(), relayUdpAddr)
-	myControl.InjectLightHouseAddr(relayVpnV6.Addr(), relayUdpAddr)
-	myControl.InjectRelays(theirVpnV6.Addr(), []netip.Addr{relayVpnV6.Addr()})
-	relayControl.InjectLightHouseAddr(theirVpnV6.Addr(), theirUdpAddr)
-
-	// Build a router so we don't have to reason who gets which packet
-	r := router.NewR(t, myControl, relayControl, theirControl)
-	defer r.RenderFlow()
-
-	// Start the servers
-	myControl.Start()
-	relayControl.Start()
-	theirControl.Start()
-
-	t.Log("Trigger a handshake from me to them via the relay")
-	myControl.InjectTunUDPPacket(theirVpnV6.Addr(), 80, myVpnV6.Addr(), 80, []byte("Hi from me"))
-
-	p := r.RouteForAllUntilTxTun(theirControl)
-	r.Log("Assert the tunnel works")
-	assertUdpPacket(t, []byte("Hi from me"), p, myVpnV6.Addr(), theirVpnV6.Addr(), 80, 80)
-
-	t.Log("reply?")
-	theirControl.InjectTunUDPPacket(myVpnV6.Addr(), 80, theirVpnV6.Addr(), 80, []byte("Hi from them"))
-	p = r.RouteForAllUntilTxTun(myControl)
-	assertUdpPacket(t, []byte("Hi from them"), p, theirVpnV6.Addr(), myVpnV6.Addr(), 80, 80)
-
-	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
-	//t.Log("finish up")
-	//myControl.Stop()
-	//theirControl.Stop()
-	//relayControl.Stop()
-}

examples/config.yml

@@ -132,6 +132,13 @@ listen:
   # Sets the max number of packets to pull from the kernel for each syscall (under systems that support recvmmsg)
   # default is 64, does not support reload
   #batch: 64
+
+# Control batching between UDP and TUN pipelines
+#batch:
+#  inbound_size: 32        # packets to queue from UDP before handing to workers
+#  outbound_size: 32       # packets to queue from TUN before handing to workers
+#  flush_interval: 50us    # flush partially filled batches after this duration
+#  max_outstanding: 1028   # batches buffered per routine on each channel
   # Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
   # Default is net.core.rmem_default and net.core.wmem_default (/proc/sys/net/core/rmem_default and /proc/sys/net/core/rmem_default)
   # Maximum is limited by memory in the system, SO_RCVBUFFORCE and SO_SNDBUFFORCE is used to avoid having to raise the system wide

handshake_ix.go

@@ -23,17 +23,13 @@ func ixHandshakeStage0(f *Interface, hh *HandshakeHostInfo) bool {
 		return false
 	}
 
+	// If we're connecting to a v6 address we must use a v2 cert
 	cs := f.pki.getCertState()
 	v := cs.initiatingVersion
-	if hh.initiatingVersionOverride != cert.VersionPre1 {
-		v = hh.initiatingVersionOverride
-	} else if v < cert.Version2 {
-		// If we're connecting to a v6 address we must use a v2 cert
-		for _, a := range hh.hostinfo.vpnAddrs {
-			if a.Is6() {
-				v = cert.Version2
-				break
-			}
+	for _, a := range hh.hostinfo.vpnAddrs {
+		if a.Is6() {
+			v = cert.Version2
+			break
 		}
 	}
 
@@ -167,19 +163,16 @@ func ixHandshakeStage1(f *Interface, addr netip.AddrPort, via *ViaSender, packet
 
 	if remoteCert.Certificate.Version() != ci.myCert.Version() {
 		// We started off using the wrong certificate version, lets see if we can match the version that was sent to us
-		myCertOtherVersion := cs.getCertificate(remoteCert.Certificate.Version())
-		if myCertOtherVersion == nil {
-			if f.l.Level >= logrus.DebugLevel {
-				f.l.WithError(err).WithFields(m{
-					"udpAddr":   addr,
-					"handshake": m{"stage": 1, "style": "ix_psk0"},
-					"cert":      remoteCert,
-				}).Debug("Might be unable to handshake with host due to missing certificate version")
-			}
-		} else {
-			// Record the certificate we are actually using
-			ci.myCert = myCertOtherVersion
+		rc := cs.getCertificate(remoteCert.Certificate.Version())
+		if rc == nil {
+			f.l.WithError(err).WithField("udpAddr", addr).
+				WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).WithField("cert", remoteCert).
+				Info("Unable to handshake with host due to missing certificate version")
+			return
 		}
+
+		// Record the certificate we are actually using
+		ci.myCert = rc
 	}
 
 	if len(remoteCert.Certificate.Networks()) == 0 {

handshake_manager.go

@@ -68,12 +68,11 @@ type HandshakeManager struct {
 type HandshakeHostInfo struct {
 	sync.Mutex
 
-	startTime                 time.Time        // Time that we first started trying with this handshake
-	ready                     bool             // Is the handshake ready
-	initiatingVersionOverride cert.Version     // Should we use a non-default cert version for this handshake?
-	counter                   int64            // How many attempts have we made so far
-	lastRemotes               []netip.AddrPort // Remotes that we sent to during the previous attempt
-	packetStore               []*cachedPacket  // A set of packets to be transmitted once the handshake completes
+	startTime   time.Time        // Time that we first started trying with this handshake
+	ready       bool             // Is the handshake ready
+	counter     int64            // How many attempts have we made so far
+	lastRemotes []netip.AddrPort // Remotes that we sent to during the previous attempt
+	packetStore []*cachedPacket  // A set of packets to be transmitted once the handshake completes
 
 	hostinfo *HostInfo
 }
@@ -300,8 +299,6 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 						InitiatorRelayIndex: idx,
 					}
 
-					relayFrom := hm.f.myVpnAddrs[0]
-
 					switch relayHostInfo.GetCert().Certificate.Version() {
 					case cert.Version1:
 						if !hm.f.myVpnAddrs[0].Is4() {
@@ -319,13 +316,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 						b = vpnIp.As4()
 						m.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 					case cert.Version2:
-						if vpnIp.Is4() {
-							relayFrom = hm.f.myVpnAddrs[0]
-						} else {
-							//todo do this smarter
-							relayFrom = hm.f.myVpnAddrs[len(hm.f.myVpnAddrs)-1]
-						}
-						m.RelayFromAddr = netAddrToProtoAddr(relayFrom)
+						m.RelayFromAddr = netAddrToProtoAddr(hm.f.myVpnAddrs[0])
 						m.RelayToAddr = netAddrToProtoAddr(vpnIp)
 					default:
 						hostinfo.logger(hm.l).Error("Unknown certificate version found while creating relay")
@@ -340,7 +331,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					} else {
 						hm.f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
 						hm.l.WithFields(logrus.Fields{
-							"relayFrom":           relayFrom,
+							"relayFrom":           hm.f.myVpnAddrs[0],
 							"relayTo":             vpnIp,
 							"initiatorRelayIndex": idx,
 							"relay":               relay}).
@@ -366,8 +357,6 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					InitiatorRelayIndex: existingRelay.LocalIndex,
 				}
 
-				relayFrom := hm.f.myVpnAddrs[0]
-
 				switch relayHostInfo.GetCert().Certificate.Version() {
 				case cert.Version1:
 					if !hm.f.myVpnAddrs[0].Is4() {
@@ -385,14 +374,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					b = vpnIp.As4()
 					m.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 				case cert.Version2:
-					if vpnIp.Is4() {
-						relayFrom = hm.f.myVpnAddrs[0]
-					} else {
-						//todo do this smarter
-						relayFrom = hm.f.myVpnAddrs[len(hm.f.myVpnAddrs)-1]
-					}
-
-					m.RelayFromAddr = netAddrToProtoAddr(relayFrom)
+					m.RelayFromAddr = netAddrToProtoAddr(hm.f.myVpnAddrs[0])
 					m.RelayToAddr = netAddrToProtoAddr(vpnIp)
 				default:
 					hostinfo.logger(hm.l).Error("Unknown certificate version found while creating relay")
@@ -407,7 +389,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
 					// This must send over the hostinfo, not over hm.Hosts[ip]
 					hm.f.SendMessageToHostInfo(header.Control, 0, relayHostInfo, msg, make([]byte, 12), make([]byte, mtu))
 					hm.l.WithFields(logrus.Fields{
-						"relayFrom":           relayFrom,
+						"relayFrom":           hm.f.myVpnAddrs[0],
 						"relayTo":             vpnIp,
 						"initiatorRelayIndex": existingRelay.LocalIndex,
 						"relay":               relay}).

hostmap.go

@@ -2,7 +2,6 @@ package nebula
 
 import (
 	"errors"
-	"fmt"
 	"net"
 	"net/netip"
 	"slices"
@@ -522,7 +521,6 @@ func (hm *HostMap) QueryVpnAddrsRelayFor(targetIps []netip.Addr, relayHostIp net
 		return nil, nil, errors.New("unable to find host")
 	}
 
-	lastH := h
 	for h != nil {
 		for _, targetIp := range targetIps {
 			r, ok := h.relayState.QueryRelayForByIp(targetIp)
@@ -530,12 +528,10 @@ func (hm *HostMap) QueryVpnAddrsRelayFor(targetIps []netip.Addr, relayHostIp net
 				return h, r, nil
 			}
 		}
-		lastH = h
 		h = h.next
 	}
 
-	//todo no merge
-	return nil, nil, fmt.Errorf("unable to find host with relay: %v", lastH)
+	return nil, nil, errors.New("unable to find host with relay")
 }
 
 func (hm *HostMap) unlockedDisestablishVpnAddrRelayFor(hi *HostInfo) {

interface.go

@@ -6,8 +6,8 @@ import (
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
 	"runtime"
+	"sync"
 	"sync/atomic"
 	"time"
 
@@ -18,10 +18,18 @@ import (
 	"github.com/slackhq/nebula/firewall"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/overlay"
+	"github.com/slackhq/nebula/packet"
 	"github.com/slackhq/nebula/udp"
 )
 
-const mtu = 9001
+const (
+	mtu = 9001
+
+	inboundBatchSizeDefault      = 32
+	outboundBatchSizeDefault     = 32
+	batchFlushIntervalDefault    = 50 * time.Microsecond
+	maxOutstandingBatchesDefault = 1028
+)
 
 type InterfaceConfig struct {
 	HostMap            *HostMap
@@ -47,9 +55,17 @@ type InterfaceConfig struct {
 	reQueryWait     time.Duration
 
 	ConntrackCacheTimeout time.Duration
+	BatchConfig           BatchConfig
 	l                     *logrus.Logger
 }
 
+type BatchConfig struct {
+	InboundBatchSize      int
+	OutboundBatchSize     int
+	FlushInterval         time.Duration
+	MaxOutstandingPerChan int
+}
+
 type Interface struct {
 	hostMap               *HostMap
 	outside               udp.Conn
@@ -87,12 +103,95 @@ type Interface struct {
 
 	writers []udp.Conn
 	readers []io.ReadWriteCloser
+	wg      sync.WaitGroup
 
 	metricHandshakes    metrics.Histogram
 	messageMetrics      *MessageMetrics
 	cachedPacketMetrics *cachedPacketMetrics
 
 	l *logrus.Logger
+
+	inPool  sync.Pool
+	inbound []chan *packetBatch
+
+	outPool  sync.Pool
+	outbound []chan *outboundBatch
+
+	packetBatchPool   sync.Pool
+	outboundBatchPool sync.Pool
+
+	inboundBatchSize      int
+	outboundBatchSize     int
+	batchFlushInterval    time.Duration
+	maxOutstandingPerChan int
+}
+
+type packetBatch struct {
+	packets []*packet.Packet
+}
+
+func newPacketBatch(capacity int) *packetBatch {
+	return &packetBatch{
+		packets: make([]*packet.Packet, 0, capacity),
+	}
+}
+
+func (b *packetBatch) add(p *packet.Packet) {
+	b.packets = append(b.packets, p)
+}
+
+func (b *packetBatch) reset() {
+	for i := range b.packets {
+		b.packets[i] = nil
+	}
+	b.packets = b.packets[:0]
+}
+
+func (f *Interface) getPacketBatch() *packetBatch {
+	if v := f.packetBatchPool.Get(); v != nil {
+		b := v.(*packetBatch)
+		b.reset()
+		return b
+	}
+	return newPacketBatch(f.inboundBatchSize)
+}
+
+func (f *Interface) releasePacketBatch(b *packetBatch) {
+	b.reset()
+	f.packetBatchPool.Put(b)
+}
+
+type outboundBatch struct {
+	payloads []*[]byte
+}
+
+func newOutboundBatch(capacity int) *outboundBatch {
+	return &outboundBatch{payloads: make([]*[]byte, 0, capacity)}
+}
+
+func (b *outboundBatch) add(buf *[]byte) {
+	b.payloads = append(b.payloads, buf)
+}
+
+func (b *outboundBatch) reset() {
+	for i := range b.payloads {
+		b.payloads[i] = nil
+	}
+	b.payloads = b.payloads[:0]
+}
+
+func (f *Interface) getOutboundBatch() *outboundBatch {
+	if v := f.outboundBatchPool.Get(); v != nil {
+		b := v.(*outboundBatch)
+		b.reset()
+		return b
+	}
+	return newOutboundBatch(f.outboundBatchSize)
+}
+
+func (f *Interface) releaseOutboundBatch(b *outboundBatch) {
+	b.reset()
+	f.outboundBatchPool.Put(b)
 }
 
 type EncWriter interface {
@@ -162,6 +261,20 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 	}
 
 	cs := c.pki.getCertState()
+
+	bc := c.BatchConfig
+	if bc.InboundBatchSize <= 0 {
+		bc.InboundBatchSize = inboundBatchSizeDefault
+	}
+	if bc.OutboundBatchSize <= 0 {
+		bc.OutboundBatchSize = outboundBatchSizeDefault
+	}
+	if bc.FlushInterval <= 0 {
+		bc.FlushInterval = batchFlushIntervalDefault
+	}
+	if bc.MaxOutstandingPerChan <= 0 {
+		bc.MaxOutstandingPerChan = maxOutstandingBatchesDefault
+	}
 	ifce := &Interface{
 		pki:                   c.pki,
 		hostMap:               c.HostMap,
@@ -194,9 +307,39 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 			dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
 		},
 
+		inbound:  make([]chan *packetBatch, c.routines),
+		outbound: make([]chan *outboundBatch, c.routines),
+
 		l: c.l,
+
+		inboundBatchSize:      bc.InboundBatchSize,
+		outboundBatchSize:     bc.OutboundBatchSize,
+		batchFlushInterval:    bc.FlushInterval,
+		maxOutstandingPerChan: bc.MaxOutstandingPerChan,
 	}
 
+	for i := 0; i < c.routines; i++ {
+		ifce.inbound[i] = make(chan *packetBatch, ifce.maxOutstandingPerChan)
+		ifce.outbound[i] = make(chan *outboundBatch, ifce.maxOutstandingPerChan)
+	}
+
+	ifce.inPool = sync.Pool{New: func() any {
+		return packet.New()
+	}}
+
+	ifce.outPool = sync.Pool{New: func() any {
+		t := make([]byte, mtu)
+		return &t
+	}}
+
+	ifce.packetBatchPool = sync.Pool{New: func() any {
+		return newPacketBatch(ifce.inboundBatchSize)
+	}}
+
+	ifce.outboundBatchPool = sync.Pool{New: func() any {
+		return newOutboundBatch(ifce.outboundBatchSize)
+	}}
+
 	ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
 	ifce.reQueryEvery.Store(c.reQueryEvery)
 	ifce.reQueryWait.Store(int64(c.reQueryWait))
@@ -209,7 +352,7 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 // activate creates the interface on the host. After the interface is created, any
 // other services that want to bind listeners to its IP may do so successfully. However,
 // the interface isn't going to process anything until run() is called.
-func (f *Interface) activate() {
+func (f *Interface) activate() error {
 	// actually turn on tun dev
 
 	addr, err := f.outside.LocalAddr()
@@ -230,33 +373,44 @@ func (f *Interface) activate() {
 		if i > 0 {
 			reader, err = f.inside.NewMultiQueueReader()
 			if err != nil {
-				f.l.Fatal(err)
+				return err
 			}
 		}
 		f.readers[i] = reader
 	}
 
-	if err := f.inside.Activate(); err != nil {
+	if err = f.inside.Activate(); err != nil {
 		f.inside.Close()
-		f.l.Fatal(err)
+		return err
 	}
+
+	return nil
 }
 
-func (f *Interface) run() {
-	// Launch n queues to read packets from udp
+func (f *Interface) run(c context.Context) (func(), error) {
 	for i := 0; i < f.routines; i++ {
+		// Launch n queues to read packets from udp
+		f.wg.Add(1)
 		go f.listenOut(i)
+
+		// Launch n queues to read packets from tun dev
+		f.wg.Add(1)
+		go f.listenIn(f.readers[i], i)
+
+		// Launch n queues to read packets from tun dev
+		f.wg.Add(1)
+		go f.workerIn(i, c)
+
+		// Launch n queues to read packets from tun dev
+		f.wg.Add(1)
+		go f.workerOut(i, c)
 	}
 
-	// Launch n queues to read packets from tun dev
-	for i := 0; i < f.routines; i++ {
-		go f.listenIn(f.readers[i], i)
-	}
+	return f.wg.Wait, nil
 }
 
 func (f *Interface) listenOut(i int) {
 	runtime.LockOSThread()
-
 	var li udp.Conn
 	if i > 0 {
 		li = f.writers[i]
@@ -264,41 +418,142 @@ func (f *Interface) listenOut(i int) {
 		li = f.outside
 	}
 
-	ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
-	lhh := f.lightHouse.NewRequestHandler()
-	plaintext := make([]byte, udp.MTU)
-	h := &header.H{}
-	fwPacket := &firewall.Packet{}
-	nb := make([]byte, 12, 12)
+	batch := f.getPacketBatch()
+	lastFlush := time.Now()
 
-	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
-		f.readOutsidePackets(fromUdpAddr, nil, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
+	flush := func(force bool) {
+		if len(batch.packets) == 0 {
+			if force {
+				f.releasePacketBatch(batch)
+			}
+			return
+		}
+
+		f.inbound[i] <- batch
+		batch = f.getPacketBatch()
+		lastFlush = time.Now()
+	}
+
+	err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
+		p := f.inPool.Get().(*packet.Packet)
+		p.Payload = p.Payload[:mtu]
+		copy(p.Payload, payload)
+		p.Payload = p.Payload[:len(payload)]
+		p.Addr = fromUdpAddr
+		batch.add(p)
+
+		if len(batch.packets) >= f.inboundBatchSize || time.Since(lastFlush) >= f.batchFlushInterval {
+			flush(false)
+		}
 	})
+
+	if len(batch.packets) > 0 {
+		f.inbound[i] <- batch
+	} else {
+		f.releasePacketBatch(batch)
+	}
+
+	if err != nil && !f.closed.Load() {
+		f.l.WithError(err).Error("Error while reading packet inbound packet, closing")
+		//TODO: Trigger Control to close
+	}
+
+	f.l.Debugf("underlay reader %v is done", i)
+	f.wg.Done()
 }
 
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
-	packet := make([]byte, mtu)
-	out := make([]byte, mtu)
-	fwPacket := &firewall.Packet{}
-	nb := make([]byte, 12, 12)
+	batch := f.getOutboundBatch()
+	lastFlush := time.Now()
 
-	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
-
-	for {
-		n, err := reader.Read(packet)
-		if err != nil {
-			if errors.Is(err, os.ErrClosed) && f.closed.Load() {
-				return
+	flush := func(force bool) {
+		if len(batch.payloads) == 0 {
+			if force {
+				f.releaseOutboundBatch(batch)
 			}
-
-			f.l.WithError(err).Error("Error while reading outbound packet")
-			// This only seems to happen when something fatal happens to the fd, so exit.
-			os.Exit(2)
+			return
 		}
 
-		f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
+		f.outbound[i] <- batch
+		batch = f.getOutboundBatch()
+		lastFlush = time.Now()
+	}
+
+	for {
+		p := f.outPool.Get().(*[]byte)
+		*p = (*p)[:mtu]
+		n, err := reader.Read(*p)
+		if err != nil {
+			if !f.closed.Load() {
+				f.l.WithError(err).Error("Error while reading outbound packet, closing")
+				//TODO: Trigger Control to close
+			}
+			break
+		}
+
+		*p = (*p)[:n]
+		batch.add(p)
+
+		if len(batch.payloads) >= f.outboundBatchSize || time.Since(lastFlush) >= f.batchFlushInterval {
+			flush(false)
+		}
+	}
+
+	if len(batch.payloads) > 0 {
+		f.outbound[i] <- batch
+	} else {
+		f.releaseOutboundBatch(batch)
+	}
+
+	f.l.Debugf("overlay reader %v is done", i)
+	f.wg.Done()
+}
+
+func (f *Interface) workerIn(i int, ctx context.Context) {
+	lhh := f.lightHouse.NewRequestHandler()
+	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+	fwPacket2 := &firewall.Packet{}
+	nb2 := make([]byte, 12, 12)
+	result2 := make([]byte, mtu)
+	h := &header.H{}
+
+	for {
+		select {
+		case batch := <-f.inbound[i]:
+			for _, p := range batch.packets {
+				f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload, h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
+				p.Payload = p.Payload[:mtu]
+				f.inPool.Put(p)
+			}
+			f.releasePacketBatch(batch)
+		case <-ctx.Done():
+			f.wg.Done()
+			return
+		}
+	}
+}
+
+func (f *Interface) workerOut(i int, ctx context.Context) {
+	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
+	fwPacket1 := &firewall.Packet{}
+	nb1 := make([]byte, 12, 12)
+	result1 := make([]byte, mtu)
+
+	for {
+		select {
+		case batch := <-f.outbound[i]:
+			for _, data := range batch.payloads {
+				f.consumeInsidePacket(*data, fwPacket1, nb1, result1, i, conntrackCache.Get(f.l))
+				*data = (*data)[:mtu]
+				f.outPool.Put(data)
+			}
+			f.releaseOutboundBatch(batch)
+		case <-ctx.Done():
+			f.wg.Done()
+			return
+		}
 	}
 }
 
@@ -451,6 +706,7 @@ func (f *Interface) GetCertState() *CertState {
 func (f *Interface) Close() error {
 	f.closed.Store(true)
 
+	// Release the udp readers
 	for _, u := range f.writers {
 		err := u.Close()
 		if err != nil {
@@ -458,6 +714,13 @@ func (f *Interface) Close() error {
 		}
 	}
 
-	// Release the tun device
-	return f.inside.Close()
+	// Release the tun readers
+	for _, u := range f.readers {
+		err := u.Close()
+		if err != nil {
+			f.l.WithError(err).Error("Error while closing tun device")
+		}
+	}
+
+	return nil
 }

main.go

@@ -221,6 +221,13 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		}
 	}
 
+	batchCfg := BatchConfig{
+		InboundBatchSize:      c.GetInt("batch.inbound_size", inboundBatchSizeDefault),
+		OutboundBatchSize:     c.GetInt("batch.outbound_size", outboundBatchSizeDefault),
+		FlushInterval:         c.GetDuration("batch.flush_interval", batchFlushIntervalDefault),
+		MaxOutstandingPerChan: c.GetInt("batch.max_outstanding", maxOutstandingBatchesDefault),
+	}
+
 	ifConfig := &InterfaceConfig{
 		HostMap:               hostMap,
 		Inside:                tun,
@@ -242,6 +249,7 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		relayManager:          NewRelayManager(ctx, l, hostMap, c),
 		punchy:                punchy,
 		ConntrackCacheTimeout: conntrackCacheTimeout,
+		BatchConfig:           batchCfg,
 		l:                     l,
 	}
 
@@ -284,14 +292,14 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 	}
 
 	return &Control{
-		ifce,
-		l,
-		ctx,
-		cancel,
-		sshStart,
-		statsStart,
-		dnsStart,
-		lightHouse.StartUpdateWorker,
-		connManager.Start,
+		f:                      ifce,
+		l:                      l,
+		ctx:                    ctx,
+		cancel:                 cancel,
+		sshStart:               sshStart,
+		statsStart:             statsStart,
+		dnsStart:               dnsStart,
+		lighthouseStart:        lightHouse.StartUpdateWorker,
+		connectionManagerStart: connManager.Start,
 	}, nil
 }

outside.go

@@ -29,7 +29,7 @@ func (f *Interface) readOutsidePackets(ip netip.AddrPort, via *ViaSender, out []
 		return
 	}
 
-	//l.Error("in packet ", header, packet[HeaderLen:])
+	//f.l.Error("in packet ", h)
 	if ip.IsValid() {
 		if f.myVpnNetworksTable.Contains(ip.Addr()) {
 			if f.l.Level >= logrus.DebugLevel {
@@ -245,6 +245,7 @@ func (f *Interface) handleHostRoaming(hostinfo *HostInfo, udpAddr netip.AddrPort
 			return
 		}
 
+		//TODO: Seems we have a bunch of stuff racing here, since we don't have a lock on hostinfo anymore we announce roaming in bursts
 		hostinfo.logger(f.l).WithField("udpAddr", hostinfo.remote).WithField("newAddr", udpAddr).
 			Info("Host roamed to new udp ip/port.")
 		hostinfo.lastRoam = time.Now()
@@ -470,7 +471,7 @@ func (f *Interface) decryptToTun(hostinfo *HostInfo, messageCounter uint64, out
 
 	out, err = hostinfo.ConnectionState.dKey.DecryptDanger(out, packet[:header.Len], packet[header.Len:], messageCounter, nb)
 	if err != nil {
-		hostinfo.logger(f.l).WithError(err).Error("Failed to decrypt packet")
+		hostinfo.logger(f.l).WithError(err).WithField("fwPacket", fwPacket).Error("Failed to decrypt packet")
 		return false
 	}
 

overlay/tun.go

@@ -1,6 +1,7 @@
 package overlay
 
 import (
+	"net"
 	"net/netip"
 
 	"github.com/sirupsen/logrus"
@@ -70,3 +71,13 @@ func findRemovedRoutes(newRoutes, oldRoutes []Route) []Route {
 
 	return removed
 }
+
+func prefixToMask(prefix netip.Prefix) netip.Addr {
+	pLen := 128
+	if prefix.Addr().Is4() {
+		pLen = 32
+	}
+
+	addr, _ := netip.AddrFromSlice(net.CIDRMask(prefix.Bits(), pLen))
+	return addr
+}

overlay/tun_darwin.go

@@ -7,7 +7,6 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"net"
 	"net/netip"
 	"os"
 	"sync/atomic"
@@ -554,13 +553,3 @@ func (t *tun) Name() string {
 func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for darwin")
 }
-
-func prefixToMask(prefix netip.Prefix) netip.Addr {
-	pLen := 128
-	if prefix.Addr().Is4() {
-		pLen = 32
-	}
-
-	addr, _ := netip.AddrFromSlice(net.CIDRMask(prefix.Bits(), pLen))
-	return addr
-}

overlay/tun_freebsd.go

@@ -10,11 +10,9 @@ import (
 	"io"
 	"io/fs"
 	"net/netip"
-	"os"
-	"os/exec"
-	"strconv"
 	"sync/atomic"
 	"syscall"
+	"time"
 	"unsafe"
 
 	"github.com/gaissmai/bart"
@@ -22,12 +20,18 @@ import (
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/routing"
 	"github.com/slackhq/nebula/util"
+	netroute "golang.org/x/net/route"
+	"golang.org/x/sys/unix"
 )
 
 const (
 	// FIODGNAME is defined in sys/sys/filio.h on FreeBSD
 	// For 32-bit systems, use FIODGNAME_32 (not defined in this file: 0x80086678)
-	FIODGNAME = 0x80106678
+	FIODGNAME        = 0x80106678
+	TUNSIFMODE       = 0x8004745e
+	TUNSIFHEAD       = 0x80047460
+	OSIOCAIFADDR_IN6 = 0x8088691b
+	IN6_IFF_NODAD    = 0x0020
 )
 
 type fiodgnameArg struct {
@@ -37,43 +41,159 @@ type fiodgnameArg struct {
 }
 
 type ifreqRename struct {
-	Name [16]byte
+	Name [unix.IFNAMSIZ]byte
 	Data uintptr
 }
 
 type ifreqDestroy struct {
-	Name [16]byte
+	Name [unix.IFNAMSIZ]byte
 	pad  [16]byte
 }
 
+type ifReq struct {
+	Name  [unix.IFNAMSIZ]byte
+	Flags uint16
+}
+
+type ifreqMTU struct {
+	Name [unix.IFNAMSIZ]byte
+	MTU  int32
+}
+
+type addrLifetime struct {
+	Expire    uint64
+	Preferred uint64
+	Vltime    uint32
+	Pltime    uint32
+}
+
+type ifreqAlias4 struct {
+	Name     [unix.IFNAMSIZ]byte
+	Addr     unix.RawSockaddrInet4
+	DstAddr  unix.RawSockaddrInet4
+	MaskAddr unix.RawSockaddrInet4
+	VHid     uint32
+}
+
+type ifreqAlias6 struct {
+	Name       [unix.IFNAMSIZ]byte
+	Addr       unix.RawSockaddrInet6
+	DstAddr    unix.RawSockaddrInet6
+	PrefixMask unix.RawSockaddrInet6
+	Flags      uint32
+	Lifetime   addrLifetime
+	VHid       uint32
+}
+
 type tun struct {
 	Device      string
 	vpnNetworks []netip.Prefix
 	MTU         int
 	Routes      atomic.Pointer[[]Route]
 	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	linkAddr    *netroute.LinkAddr
 	l           *logrus.Logger
+	devFd       int
+}
 
-	io.ReadWriteCloser
+func (t *tun) Read(to []byte) (int, error) {
+	// use readv() to read from the tunnel device, to eliminate the need for copying the buffer
+	if t.devFd < 0 {
+		return -1, syscall.EINVAL
+	}
+
+	// first 4 bytes is protocol family, in network byte order
+	head := make([]byte, 4)
+
+	iovecs := []syscall.Iovec{
+		{&head[0], 4},
+		{&to[0], uint64(len(to))},
+	}
+
+	n, _, errno := syscall.Syscall(syscall.SYS_READV, uintptr(t.devFd), uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
+
+	var err error
+	if errno != 0 {
+		err = syscall.Errno(errno)
+	} else {
+		err = nil
+	}
+	// fix bytes read number to exclude header
+	bytesRead := int(n)
+	if bytesRead < 0 {
+		return bytesRead, err
+	} else if bytesRead < 4 {
+		return 0, err
+	} else {
+		return bytesRead - 4, err
+	}
+}
+
+// Write is only valid for single threaded use
+func (t *tun) Write(from []byte) (int, error) {
+	// use writev() to write to the tunnel device, to eliminate the need for copying the buffer
+	if t.devFd < 0 {
+		return -1, syscall.EINVAL
+	}
+
+	if len(from) <= 1 {
+		return 0, syscall.EIO
+	}
+	ipVer := from[0] >> 4
+	var head []byte
+	// first 4 bytes is protocol family, in network byte order
+	if ipVer == 4 {
+		head = []byte{0, 0, 0, syscall.AF_INET}
+	} else if ipVer == 6 {
+		head = []byte{0, 0, 0, syscall.AF_INET6}
+	} else {
+		return 0, fmt.Errorf("unable to determine IP version from packet")
+	}
+	iovecs := []syscall.Iovec{
+		{&head[0], 4},
+		{&from[0], uint64(len(from))},
+	}
+
+	n, _, errno := syscall.Syscall(syscall.SYS_WRITEV, uintptr(t.devFd), uintptr(unsafe.Pointer(&iovecs[0])), uintptr(2))
+
+	var err error
+	if errno != 0 {
+		err = syscall.Errno(errno)
+	} else {
+		err = nil
+	}
+
+	return int(n) - 4, err
 }
 
 func (t *tun) Close() error {
-	if t.ReadWriteCloser != nil {
-		if err := t.ReadWriteCloser.Close(); err != nil {
-			return err
-		}
-
-		s, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
+	if t.devFd >= 0 {
+		err := syscall.Close(t.devFd)
 		if err != nil {
-			return err
+			t.l.WithError(err).Error("Error closing device")
 		}
-		defer syscall.Close(s)
+		t.devFd = -1
 
-		ifreq := ifreqDestroy{Name: t.deviceBytes()}
+		c := make(chan struct{})
+		go func() {
+			// destroying the interface can block if a read() is still pending. Do this asynchronously.
+			defer close(c)
+			s, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
+			if err == nil {
+				defer syscall.Close(s)
+				ifreq := ifreqDestroy{Name: t.deviceBytes()}
+				err = ioctl(uintptr(s), syscall.SIOCIFDESTROY, uintptr(unsafe.Pointer(&ifreq)))
+			}
+			if err != nil {
+				t.l.WithError(err).Error("Error destroying tunnel")
+			}
+		}()
 
-		// Destroy the interface
-		err = ioctl(uintptr(s), syscall.SIOCIFDESTROY, uintptr(unsafe.Pointer(&ifreq)))
-		return err
+		// wait up to 1 second so we start blocking at the ioctl
+		select {
+		case <-c:
+		case <-time.After(1 * time.Second):
+		}
 	}
 
 	return nil
@@ -85,32 +205,37 @@ func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun,
 
 func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
 	// Try to open existing tun device
-	var file *os.File
+	var fd int
 	var err error
 	deviceName := c.GetString("tun.dev", "")
 	if deviceName != "" {
-		file, err = os.OpenFile("/dev/"+deviceName, os.O_RDWR, 0)
+		fd, err = syscall.Open("/dev/"+deviceName, syscall.O_RDWR, 0)
 	}
 	if errors.Is(err, fs.ErrNotExist) || deviceName == "" {
 		// If the device doesn't already exist, request a new one and rename it
-		file, err = os.OpenFile("/dev/tun", os.O_RDWR, 0)
+		fd, err = syscall.Open("/dev/tun", syscall.O_RDWR, 0)
 	}
 	if err != nil {
 		return nil, err
 	}
 
-	rawConn, err := file.SyscallConn()
-	if err != nil {
-		return nil, fmt.Errorf("SyscallConn: %v", err)
+	// Read the name of the interface
+	var name [16]byte
+	arg := fiodgnameArg{length: 16, buf: unsafe.Pointer(&name)}
+	ctrlErr := ioctl(uintptr(fd), FIODGNAME, uintptr(unsafe.Pointer(&arg)))
+
+	if ctrlErr == nil {
+		// set broadcast mode and multicast
+		ifmode := uint32(unix.IFF_BROADCAST | unix.IFF_MULTICAST)
+		ctrlErr = ioctl(uintptr(fd), TUNSIFMODE, uintptr(unsafe.Pointer(&ifmode)))
+	}
+
+	if ctrlErr == nil {
+		// turn on link-layer mode, to support ipv6
+		ifhead := uint32(1)
+		ctrlErr = ioctl(uintptr(fd), TUNSIFHEAD, uintptr(unsafe.Pointer(&ifhead)))
 	}
 
-	var name [16]byte
-	var ctrlErr error
-	rawConn.Control(func(fd uintptr) {
-		// Read the name of the interface
-		arg := fiodgnameArg{length: 16, buf: unsafe.Pointer(&name)}
-		ctrlErr = ioctl(fd, FIODGNAME, uintptr(unsafe.Pointer(&arg)))
-	})
 	if ctrlErr != nil {
 		return nil, err
 	}
@@ -122,11 +247,7 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 
 	// If the name doesn't match the desired interface name, rename it now
 	if ifName != deviceName {
-		s, err := syscall.Socket(
-			syscall.AF_INET,
-			syscall.SOCK_DGRAM,
-			syscall.IPPROTO_IP,
-		)
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
 		if err != nil {
 			return nil, err
 		}
@@ -149,11 +270,11 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	}
 
 	t := &tun{
-		ReadWriteCloser: file,
-		Device:          deviceName,
-		vpnNetworks:     vpnNetworks,
-		MTU:             c.GetInt("tun.mtu", DefaultMTU),
-		l:               l,
+		Device:      deviceName,
+		vpnNetworks: vpnNetworks,
+		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		l:           l,
+		devFd:       fd,
 	}
 
 	err = t.reload(c, true)
@@ -172,38 +293,111 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 }
 
 func (t *tun) addIp(cidr netip.Prefix) error {
-	var err error
-	// TODO use syscalls instead of exec.Command
-	cmd := exec.Command("/sbin/ifconfig", t.Device, cidr.String(), cidr.Addr().String())
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
+	if cidr.Addr().Is4() {
+		ifr := ifreqAlias4{
+			Name: t.deviceBytes(),
+			Addr: unix.RawSockaddrInet4{
+				Len:    unix.SizeofSockaddrInet4,
+				Family: unix.AF_INET,
+				Addr:   cidr.Addr().As4(),
+			},
+			DstAddr: unix.RawSockaddrInet4{
+				Len:    unix.SizeofSockaddrInet4,
+				Family: unix.AF_INET,
+				Addr:   getBroadcast(cidr).As4(),
+			},
+			MaskAddr: unix.RawSockaddrInet4{
+				Len:    unix.SizeofSockaddrInet4,
+				Family: unix.AF_INET,
+				Addr:   prefixToMask(cidr).As4(),
+			},
+			VHid: 0,
+		}
+		s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+		// Note: unix.SIOCAIFADDR corresponds to FreeBSD's OSIOCAIFADDR
+		if err := ioctl(uintptr(s), unix.SIOCAIFADDR, uintptr(unsafe.Pointer(&ifr))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
+		}
+		return nil
 	}
 
-	cmd = exec.Command("/sbin/route", "-n", "add", "-net", cidr.String(), "-interface", t.Device)
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'route add': %s", err)
+	if cidr.Addr().Is6() {
+		ifr := ifreqAlias6{
+			Name: t.deviceBytes(),
+			Addr: unix.RawSockaddrInet6{
+				Len:    unix.SizeofSockaddrInet6,
+				Family: unix.AF_INET6,
+				Addr:   cidr.Addr().As16(),
+			},
+			PrefixMask: unix.RawSockaddrInet6{
+				Len:    unix.SizeofSockaddrInet6,
+				Family: unix.AF_INET6,
+				Addr:   prefixToMask(cidr).As16(),
+			},
+			Lifetime: addrLifetime{
+				Expire:    0,
+				Preferred: 0,
+				Vltime:    0xffffffff,
+				Pltime:    0xffffffff,
+			},
+			Flags: IN6_IFF_NODAD,
+		}
+		s, err := syscall.Socket(syscall.AF_INET6, syscall.SOCK_DGRAM, syscall.IPPROTO_IP)
+		if err != nil {
+			return err
+		}
+		defer syscall.Close(s)
+
+		if err := ioctl(uintptr(s), OSIOCAIFADDR_IN6, uintptr(unsafe.Pointer(&ifr))); err != nil {
+			return fmt.Errorf("failed to set tun address %s: %s", cidr.Addr().String(), err)
+		}
+		return nil
 	}
 
-	cmd = exec.Command("/sbin/ifconfig", t.Device, "mtu", strconv.Itoa(t.MTU))
-	t.l.Debug("command: ", cmd.String())
-	if err = cmd.Run(); err != nil {
-		return fmt.Errorf("failed to run 'ifconfig': %s", err)
-	}
-
-	// Unsafe path routes
-	return t.addRoutes(false)
+	return fmt.Errorf("unknown address type %v", cidr)
 }
 
 func (t *tun) Activate() error {
+	// Setup our default MTU
+	err := t.setMTU()
+	if err != nil {
+		return err
+	}
+
+	linkAddr, err := getLinkAddr(t.Device)
+	if err != nil {
+		return err
+	}
+	if linkAddr == nil {
+		return fmt.Errorf("unable to discover link_addr for tun interface")
+	}
+	t.linkAddr = linkAddr
+
 	for i := range t.vpnNetworks {
 		err := t.addIp(t.vpnNetworks[i])
 		if err != nil {
 			return err
 		}
 	}
-	return nil
+
+	return t.addRoutes(false)
+}
+
+func (t *tun) setMTU() error {
+	// Set the MTU on the device
+	s, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM, unix.IPPROTO_IP)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(s)
+
+	ifm := ifreqMTU{Name: t.deviceBytes(), MTU: int32(t.MTU)}
+	err = ioctl(uintptr(s), unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm)))
+	return err
 }
 
 func (t *tun) reload(c *config.C, initial bool) error {
@@ -268,15 +462,16 @@ func (t *tun) addRoutes(logErrors bool) error {
 			continue
 		}
 
-		cmd := exec.Command("/sbin/route", "-n", "add", "-net", r.Cidr.String(), "-interface", t.Device)
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
-			retErr := util.NewContextualError("failed to run 'route add' for unsafe_route", map[string]any{"route": r}, err)
+		err := addRoute(r.Cidr, t.linkAddr)
+		if err != nil {
+			retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
 			if logErrors {
 				retErr.Log(t.l)
 			} else {
 				return retErr
 			}
+		} else {
+			t.l.WithField("route", r).Info("Added route")
 		}
 	}
 
@@ -289,9 +484,8 @@ func (t *tun) removeRoutes(routes []Route) error {
 			continue
 		}
 
-		cmd := exec.Command("/sbin/route", "-n", "delete", "-net", r.Cidr.String(), "-interface", t.Device)
-		t.l.Debug("command: ", cmd.String())
-		if err := cmd.Run(); err != nil {
+		err := delRoute(r.Cidr, t.linkAddr)
+		if err != nil {
 			t.l.WithError(err).WithField("route", r).Error("Failed to remove route")
 		} else {
 			t.l.WithField("route", r).Info("Removed route")
@@ -306,3 +500,144 @@ func (t *tun) deviceBytes() (o [16]byte) {
 	}
 	return
 }
+
+func flipBytes(b []byte) []byte {
+	for i := 0; i < len(b); i++ {
+		b[i] ^= 0xFF
+	}
+	return b
+}
+func orBytes(a []byte, b []byte) []byte {
+	ret := make([]byte, len(a))
+	for i := 0; i < len(a); i++ {
+		ret[i] = a[i] | b[i]
+	}
+	return ret
+}
+
+func getBroadcast(cidr netip.Prefix) netip.Addr {
+	broadcast, _ := netip.AddrFromSlice(
+		orBytes(
+			cidr.Addr().AsSlice(),
+			flipBytes(prefixToMask(cidr).AsSlice()),
+		),
+	)
+	return broadcast
+}
+
+func addRoute(prefix netip.Prefix, gateway netroute.Addr) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := &netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_ADD,
+		Flags:   unix.RTF_UP,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: gateway,
+		}
+	} else {
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: gateway,
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		if errors.Is(err, unix.EEXIST) {
+			// Try to do a change
+			route.Type = unix.RTM_CHANGE
+			data, err = route.Marshal()
+			if err != nil {
+				return fmt.Errorf("failed to create route.RouteMessage for change: %w", err)
+			}
+			_, err = unix.Write(sock, data[:])
+			fmt.Println("DOING CHANGE")
+			return err
+		}
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}
+
+func delRoute(prefix netip.Prefix, gateway netroute.Addr) error {
+	sock, err := unix.Socket(unix.AF_ROUTE, unix.SOCK_RAW, unix.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("unable to create AF_ROUTE socket: %v", err)
+	}
+	defer unix.Close(sock)
+
+	route := netroute.RouteMessage{
+		Version: unix.RTM_VERSION,
+		Type:    unix.RTM_DELETE,
+		Seq:     1,
+	}
+
+	if prefix.Addr().Is4() {
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet4Addr{IP: prefix.Masked().Addr().As4()},
+			unix.RTAX_NETMASK: &netroute.Inet4Addr{IP: prefixToMask(prefix).As4()},
+			unix.RTAX_GATEWAY: gateway,
+		}
+	} else {
+		route.Addrs = []netroute.Addr{
+			unix.RTAX_DST:     &netroute.Inet6Addr{IP: prefix.Masked().Addr().As16()},
+			unix.RTAX_NETMASK: &netroute.Inet6Addr{IP: prefixToMask(prefix).As16()},
+			unix.RTAX_GATEWAY: gateway,
+		}
+	}
+
+	data, err := route.Marshal()
+	if err != nil {
+		return fmt.Errorf("failed to create route.RouteMessage: %w", err)
+	}
+	_, err = unix.Write(sock, data[:])
+	if err != nil {
+		return fmt.Errorf("failed to write route.RouteMessage to socket: %w", err)
+	}
+
+	return nil
+}
+
+// getLinkAddr Gets the link address for the interface of the given name
+func getLinkAddr(name string) (*netroute.LinkAddr, error) {
+	rib, err := netroute.FetchRIB(unix.AF_UNSPEC, unix.NET_RT_IFLIST, 0)
+	if err != nil {
+		return nil, err
+	}
+	msgs, err := netroute.ParseRIB(unix.NET_RT_IFLIST, rib)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, m := range msgs {
+		switch m := m.(type) {
+		case *netroute.InterfaceMessage:
+			if m.Name == name {
+				sa, ok := m.Addrs[unix.RTAX_IFP].(*netroute.LinkAddr)
+				if ok {
+					return sa, nil
+				}
+			}
+		}
+	}
+
+	return nil, nil
+}

packet/packet.go

@@ -0,0 +1,12 @@
+package packet
+
+import "net/netip"
+
+type Packet struct {
+	Payload []byte
+	Addr    netip.AddrPort
+}
+
+func New() *Packet {
+	return &Packet{Payload: make([]byte, 9001)}
+}

pki.go

@@ -100,62 +100,55 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
 		currentState := p.cs.Load()
 		if newState.v1Cert != nil {
 			if currentState.v1Cert == nil {
-				//adding certs is fine, actually. Networks-in-common confirmed in newCertState().
-			} else {
-				// did IP in cert change? if so, don't set
-				if !slices.Equal(currentState.v1Cert.Networks(), newState.v1Cert.Networks()) {
-					return util.NewContextualError(
-						"Networks in new cert was different from old",
-						m{"new_networks": newState.v1Cert.Networks(), "old_networks": currentState.v1Cert.Networks(), "cert_version": cert.Version1},
-						nil,
-					)
-				}
-
-				if currentState.v1Cert.Curve() != newState.v1Cert.Curve() {
-					return util.NewContextualError(
-						"Curve in new v1 cert was different from old",
-						m{"new_curve": newState.v1Cert.Curve(), "old_curve": currentState.v1Cert.Curve(), "cert_version": cert.Version1},
-						nil,
-					)
-				}
+				return util.NewContextualError("v1 certificate was added, restart required", nil, err)
 			}
+
+			// did IP in cert change? if so, don't set
+			if !slices.Equal(currentState.v1Cert.Networks(), newState.v1Cert.Networks()) {
+				return util.NewContextualError(
+					"Networks in new cert was different from old",
+					m{"new_networks": newState.v1Cert.Networks(), "old_networks": currentState.v1Cert.Networks()},
+					nil,
+				)
+			}
+
+			if currentState.v1Cert.Curve() != newState.v1Cert.Curve() {
+				return util.NewContextualError(
+					"Curve in new cert was different from old",
+					m{"new_curve": newState.v1Cert.Curve(), "old_curve": currentState.v1Cert.Curve()},
+					nil,
+				)
+			}
+
+		} else if currentState.v1Cert != nil {
+			//TODO: CERT-V2 we should be able to tear this down
+			return util.NewContextualError("v1 certificate was removed, restart required", nil, err)
 		}
 
 		if newState.v2Cert != nil {
 			if currentState.v2Cert == nil {
-				//adding certs is fine, actually
-			} else {
-				// did IP in cert change? if so, don't set
-				if !slices.Equal(currentState.v2Cert.Networks(), newState.v2Cert.Networks()) {
-					return util.NewContextualError(
-						"Networks in new cert was different from old",
-						m{"new_networks": newState.v2Cert.Networks(), "old_networks": currentState.v2Cert.Networks(), "cert_version": cert.Version2},
-						nil,
-					)
-				}
-
-				if currentState.v2Cert.Curve() != newState.v2Cert.Curve() {
-					return util.NewContextualError(
-						"Curve in new cert was different from old",
-						m{"new_curve": newState.v2Cert.Curve(), "old_curve": currentState.v2Cert.Curve(), "cert_version": cert.Version2},
-						nil,
-					)
-				}
+				return util.NewContextualError("v2 certificate was added, restart required", nil, err)
 			}
 
-		} else if currentState.v2Cert != nil {
-			//newState.v1Cert is non-nil bc empty certstates aren't permitted
-			if newState.v1Cert == nil {
-				return util.NewContextualError("v1 and v2 certs are nil, this should be impossible", nil, err)
-			}
-			//if we're going to v1-only, we need to make sure we didn't orphan any v2-cert vpnaddrs
-			if !slices.Equal(currentState.v2Cert.Networks(), newState.v1Cert.Networks()) {
+			// did IP in cert change? if so, don't set
+			if !slices.Equal(currentState.v2Cert.Networks(), newState.v2Cert.Networks()) {
 				return util.NewContextualError(
-					"Removing a V2 cert is not permitted unless it has identical networks to the new V1 cert",
-					m{"new_v1_networks": newState.v1Cert.Networks(), "old_v2_networks": currentState.v2Cert.Networks()},
+					"Networks in new cert was different from old",
+					m{"new_networks": newState.v2Cert.Networks(), "old_networks": currentState.v2Cert.Networks()},
 					nil,
 				)
 			}
+
+			if currentState.v2Cert.Curve() != newState.v2Cert.Curve() {
+				return util.NewContextualError(
+					"Curve in new cert was different from old",
+					m{"new_curve": newState.v2Cert.Curve(), "old_curve": currentState.v2Cert.Curve()},
+					nil,
+				)
+			}
+
+		} else if currentState.v2Cert != nil {
+			return util.NewContextualError("v2 certificate was removed, restart required", nil, err)
 		}
 
 		// Cipher cant be hot swapped so just leave it at what it was before

relay_manager.go

@@ -190,7 +190,6 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 			InitiatorRelayIndex: peerRelay.RemoteIndex,
 		}
 
-		relayFrom := h.vpnAddrs[0]
 		if v == cert.Version1 {
 			peer := peerHostInfo.vpnAddrs[0]
 			if !peer.Is4() {
@@ -208,13 +207,7 @@ func (rm *relayManager) handleCreateRelayResponse(v cert.Version, h *HostInfo, f
 			b = targetAddr.As4()
 			resp.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 		} else {
-			if targetAddr.Is4() {
-				relayFrom = h.vpnAddrs[0]
-			} else {
-				//todo do this smarter
-				relayFrom = h.vpnAddrs[len(h.vpnAddrs)-1]
-			}
-			resp.RelayFromAddr = netAddrToProtoAddr(relayFrom)
+			resp.RelayFromAddr = netAddrToProtoAddr(peerHostInfo.vpnAddrs[0])
 			resp.RelayToAddr = target
 		}
 
@@ -367,7 +360,7 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 			Type:                NebulaControl_CreateRelayRequest,
 			InitiatorRelayIndex: index,
 		}
-		relayFrom := h.vpnAddrs[0]
+
 		if v == cert.Version1 {
 			if !h.vpnAddrs[0].Is4() {
 				rm.l.WithField("relayFrom", h.vpnAddrs[0]).
@@ -384,13 +377,7 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 			b = target.As4()
 			req.OldRelayToAddr = binary.BigEndian.Uint32(b[:])
 		} else {
-			if target.Is4() {
-				relayFrom = h.vpnAddrs[0]
-			} else {
-				//todo do this smarter
-				relayFrom = h.vpnAddrs[len(h.vpnAddrs)-1]
-			}
-			req.RelayFromAddr = netAddrToProtoAddr(relayFrom)
+			req.RelayFromAddr = netAddrToProtoAddr(h.vpnAddrs[0])
 			req.RelayToAddr = netAddrToProtoAddr(target)
 		}
 
@@ -401,7 +388,7 @@ func (rm *relayManager) handleCreateRelayRequest(v cert.Version, h *HostInfo, f
 		} else {
 			f.SendMessageToHostInfo(header.Control, 0, peer, msg, make([]byte, 12), make([]byte, mtu))
 			rm.l.WithFields(logrus.Fields{
-				"relayFrom":           relayFrom,
+				"relayFrom":           h.vpnAddrs[0],
 				"relayTo":             target,
 				"initiatorRelayIndex": req.InitiatorRelayIndex,
 				"responderRelayIndex": req.ResponderRelayIndex,

service/service.go

@@ -44,7 +44,10 @@ type Service struct {
 }
 
 func New(control *nebula.Control) (*Service, error) {
-	control.Start()
+	wait, err := control.Start()
+	if err != nil {
+		return nil, err
+	}
 
 	ctx := control.Context()
 	eg, ctx := errgroup.WithContext(ctx)
@@ -141,6 +144,12 @@ func New(control *nebula.Control) (*Service, error) {
 		}
 	})
 
+	// Add the nebula wait function to the group
+	eg.Go(func() error {
+		wait()
+		return nil
+	})
+
 	return &s, nil
 }
 

udp/conn.go

@@ -16,7 +16,7 @@ type EncReader func(
 type Conn interface {
 	Rebind() error
 	LocalAddr() (netip.AddrPort, error)
-	ListenOut(r EncReader)
+	ListenOut(r EncReader) error
 	WriteTo(b []byte, addr netip.AddrPort) error
 	ReloadConfig(c *config.C)
 	Close() error
@@ -30,8 +30,8 @@ func (NoopConn) Rebind() error {
 func (NoopConn) LocalAddr() (netip.AddrPort, error) {
 	return netip.AddrPort{}, nil
 }
-func (NoopConn) ListenOut(_ EncReader) {
-	return
+func (NoopConn) ListenOut(_ EncReader) error {
+	return nil
 }
 func (NoopConn) WriteTo(_ []byte, _ netip.AddrPort) error {
 	return nil

udp/udp_darwin.go

@@ -165,7 +165,7 @@ func NewUDPStatsEmitter(udpConns []Conn) func() {
 	return func() {}
 }
 
-func (u *StdConn) ListenOut(r EncReader) {
+func (u *StdConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	for {
@@ -174,14 +174,17 @@ func (u *StdConn) ListenOut(r EncReader) {
 		if err != nil {
 			if errors.Is(err, net.ErrClosed) {
 				u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-				return
+				return err
 			}
 
 			u.l.WithError(err).Error("unexpected udp socket receive error")
+			continue
 		}
 
 		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])
 	}
+
+	return nil
 }
 
 func (u *StdConn) Rebind() error {

udp/udp_generic.go

@@ -71,15 +71,14 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *GenericConn) ListenOut(r EncReader) {
+func (u *GenericConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	for {
 		// Just read one packet at a time
 		n, rua, err := u.ReadFromUDPAddrPort(buffer)
 		if err != nil {
-			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-			return
+			return err
 		}
 
 		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])

udp/udp_linux.go

@@ -5,10 +5,13 @@ package udp
 
 import (
 	"encoding/binary"
+	"errors"
 	"fmt"
 	"net"
 	"net/netip"
+	"sync"
 	"syscall"
+	"time"
 	"unsafe"
 
 	"github.com/rcrowley/go-metrics"
@@ -17,19 +20,40 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+var readTimeout = unix.NsecToTimeval(int64(time.Millisecond * 500))
+
+const (
+	defaultGSOMaxSegments    = 8
+	defaultGSOFlushTimeout   = 150 * time.Microsecond
+	defaultGROReadBufferSize = MTU * defaultGSOMaxSegments
+	maxGSOBatchBytes         = 0xFFFF
+)
+
+var (
+	errGSOFallback = errors.New("udp gso fallback")
+	errGSODisabled = errors.New("udp gso disabled")
+)
+
 type StdConn struct {
 	sysFd int
 	isV4  bool
 	l     *logrus.Logger
 	batch int
-}
 
-func maybeIPV4(ip net.IP) (net.IP, bool) {
-	ip4 := ip.To4()
-	if ip4 != nil {
-		return ip4, true
-	}
-	return ip, false
+	enableGRO bool
+	enableGSO bool
+
+	gsoMu           sync.Mutex
+	gsoBuf          []byte
+	gsoAddr         netip.AddrPort
+	gsoSegSize      int
+	gsoSegments     int
+	gsoMaxSegments  int
+	gsoMaxBytes     int
+	gsoFlushTimeout time.Duration
+	gsoTimer        *time.Timer
+
+	groBufSize int
 }
 
 func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
@@ -55,6 +79,11 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		}
 	}
 
+	// Set a read timeout
+	if err = unix.SetsockoptTimeval(fd, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &readTimeout); err != nil {
+		return nil, fmt.Errorf("unable to set SO_RCVTIMEO: %s", err)
+	}
+
 	var sa unix.Sockaddr
 	if ip.Is4() {
 		sa4 := &unix.SockaddrInet4{Port: port}
@@ -69,7 +98,16 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		return nil, fmt.Errorf("unable to bind to socket: %s", err)
 	}
 
-	return &StdConn{sysFd: fd, isV4: ip.Is4(), l: l, batch: batch}, err
+	return &StdConn{
+		sysFd:           fd,
+		isV4:            ip.Is4(),
+		l:               l,
+		batch:           batch,
+		gsoMaxSegments:  defaultGSOMaxSegments,
+		gsoMaxBytes:     MTU * defaultGSOMaxSegments,
+		gsoFlushTimeout: defaultGSOFlushTimeout,
+		groBufSize:      MTU,
+	}, err
 }
 
 func (u *StdConn) Rebind() error {
@@ -118,20 +156,46 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 	}
 }
 
-func (u *StdConn) ListenOut(r EncReader) {
-	var ip netip.Addr
+func (u *StdConn) ListenOut(r EncReader) error {
+	var (
+		ip       netip.Addr
+		controls [][]byte
+	)
 
-	msgs, buffers, names := u.PrepareRawMessages(u.batch)
+	bufSize := u.readBufferSize()
+	msgs, buffers, names := u.PrepareRawMessages(u.batch, bufSize)
 	read := u.ReadMulti
 	if u.batch == 1 {
 		read = u.ReadSingle
 	}
 
 	for {
+		desired := u.readBufferSize()
+		if len(buffers) == 0 || cap(buffers[0]) < desired {
+			msgs, buffers, names = u.PrepareRawMessages(u.batch, desired)
+			controls = nil
+		}
+
+		if u.enableGRO {
+			if controls == nil {
+				controls = make([][]byte, len(msgs))
+				for i := range controls {
+					controls[i] = make([]byte, unix.CmsgSpace(4))
+				}
+			}
+			for i := range msgs {
+				setRawMessageControl(&msgs[i], controls[i])
+			}
+		} else if controls != nil {
+			for i := range msgs {
+				setRawMessageControl(&msgs[i], nil)
+			}
+			controls = nil
+		}
+
 		n, err := read(msgs)
 		if err != nil {
-			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-			return
+			return err
 		}
 
 		for i := 0; i < n; i++ {
@@ -141,11 +205,82 @@ func (u *StdConn) ListenOut(r EncReader) {
 			} else {
 				ip, _ = netip.AddrFromSlice(names[i][8:24])
 			}
-			r(netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4])), buffers[i][:msgs[i].Len])
+			addr := netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4]))
+			payload := buffers[i][:msgs[i].Len]
+
+			if u.enableGRO && u.l.IsLevelEnabled(logrus.DebugLevel) {
+				ctrlLen := getRawMessageControlLen(&msgs[i])
+				msgFlags := getRawMessageFlags(&msgs[i])
+				u.l.WithFields(logrus.Fields{
+					"tag":         "gro-debug",
+					"stage":       "recv",
+					"payload_len": len(payload),
+					"ctrl_len":    ctrlLen,
+					"msg_flags":   msgFlags,
+				}).Debug("gro batch data")
+				if controls != nil && ctrlLen > 0 {
+					maxDump := ctrlLen
+					if maxDump > 16 {
+						maxDump = 16
+					}
+					u.l.WithFields(logrus.Fields{
+						"tag":         "gro-debug",
+						"stage":       "control-bytes",
+						"control_hex": fmt.Sprintf("%x", controls[i][:maxDump]),
+						"datalen":     ctrlLen,
+					}).Debug("gro control dump")
+				}
+			}
+
+			sawControl := false
+			if controls != nil {
+				if ctrlLen := getRawMessageControlLen(&msgs[i]); ctrlLen > 0 {
+					if segSize, segCount := parseGROControl(controls[i][:ctrlLen]); segSize > 0 {
+						sawControl = true
+						if u.l.IsLevelEnabled(logrus.DebugLevel) {
+							u.l.WithFields(logrus.Fields{
+								"tag":        "gro-debug",
+								"stage":      "control",
+								"seg_size":   segSize,
+								"seg_count":  segCount,
+								"payloadLen": len(payload),
+							}).Debug("gro control parsed")
+						}
+						segSize = normalizeGROSegSize(segSize, segCount, len(payload))
+						if segSize > 0 && segSize < len(payload) {
+							if u.emitGROSegments(r, addr, payload, segSize) {
+								continue
+							}
+						}
+					}
+				}
+			}
+
+			if u.enableGRO && len(payload) > MTU {
+				if !sawControl && u.l.IsLevelEnabled(logrus.DebugLevel) {
+					u.l.WithFields(logrus.Fields{
+						"tag":         "gro-debug",
+						"stage":       "fallback",
+						"payload_len": len(payload),
+					}).Debug("gro control missing; splitting payload by MTU")
+				}
+				if u.emitGROSegments(r, addr, payload, MTU) {
+					continue
+				}
+			}
+
+			r(addr, payload)
 		}
 	}
 }
 
+func (u *StdConn) readBufferSize() int {
+	if u.enableGRO && u.groBufSize > MTU {
+		return u.groBufSize
+	}
+	return MTU
+}
+
 func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 	for {
 		n, _, err := unix.Syscall6(
@@ -159,6 +294,9 @@ func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 		)
 
 		if err != 0 {
+			if err == unix.EAGAIN || err == unix.EINTR {
+				continue
+			}
 			return 0, &net.OpError{Op: "recvmsg", Err: err}
 		}
 
@@ -180,6 +318,9 @@ func (u *StdConn) ReadMulti(msgs []rawMessage) (int, error) {
 		)
 
 		if err != 0 {
+			if err == unix.EAGAIN || err == unix.EINTR {
+				continue
+			}
 			return 0, &net.OpError{Op: "recvmmsg", Err: err}
 		}
 
@@ -188,6 +329,14 @@ func (u *StdConn) ReadMulti(msgs []rawMessage) (int, error) {
 }
 
 func (u *StdConn) WriteTo(b []byte, ip netip.AddrPort) error {
+	if u.enableGSO && ip.IsValid() {
+		if err := u.queueGSOPacket(b, ip); err == nil {
+			return nil
+		} else if !errors.Is(err, errGSOFallback) {
+			return err
+		}
+	}
+
 	if u.isV4 {
 		return u.writeTo4(b, ip)
 	}
@@ -221,7 +370,7 @@ func (u *StdConn) writeTo6(b []byte, ip netip.AddrPort) error {
 
 func (u *StdConn) writeTo4(b []byte, ip netip.AddrPort) error {
 	if !ip.Addr().Is4() {
-		return ErrInvalidIPv6RemoteForSocket
+		return fmt.Errorf("Listener is IPv4, but writing to IPv6 remote")
 	}
 
 	var rsa unix.RawSockaddrInet4
@@ -294,6 +443,94 @@ func (u *StdConn) ReloadConfig(c *config.C) {
 			u.l.WithError(err).Error("Failed to set listen.so_mark")
 		}
 	}
+
+	u.configureGRO(c)
+	u.configureGSO(c)
+}
+
+func (u *StdConn) configureGRO(c *config.C) {
+	if c == nil {
+		return
+	}
+
+	enable := c.GetBool("listen.enable_gro", false)
+	if enable == u.enableGRO {
+		if enable {
+			if size := c.GetInt("listen.gro_read_buffer", 0); size > 0 {
+				u.setGROBufferSize(size)
+			}
+		}
+		return
+	}
+
+	if enable {
+		if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 1); err != nil {
+			u.l.WithError(err).Warn("Failed to enable UDP GRO")
+			return
+		}
+		u.enableGRO = true
+		u.setGROBufferSize(c.GetInt("listen.gro_read_buffer", defaultGROReadBufferSize))
+		u.l.WithField("buffer_size", u.groBufSize).Info("UDP GRO enabled")
+		return
+	}
+
+	if err := unix.SetsockoptInt(u.sysFd, unix.SOL_UDP, unix.UDP_GRO, 0); err != nil && err != unix.ENOPROTOOPT {
+		u.l.WithError(err).Warn("Failed to disable UDP GRO")
+	}
+	u.enableGRO = false
+	u.groBufSize = MTU
+}
+
+func (u *StdConn) configureGSO(c *config.C) {
+	enable := c.GetBool("listen.enable_gso", false)
+	if !enable {
+		u.disableGSO()
+	} else {
+		u.enableGSO = true
+	}
+
+	segments := c.GetInt("listen.gso_max_segments", defaultGSOMaxSegments)
+	if segments < 1 {
+		segments = 1
+	}
+	u.gsoMaxSegments = segments
+
+	maxBytes := c.GetInt("listen.gso_max_bytes", 0)
+	if maxBytes <= 0 {
+		maxBytes = MTU * segments
+	}
+	if maxBytes > maxGSOBatchBytes {
+		u.l.WithField("requested", maxBytes).Warn("listen.gso_max_bytes larger than UDP limit; clamping")
+		maxBytes = maxGSOBatchBytes
+	}
+	u.gsoMaxBytes = maxBytes
+
+	timeout := c.GetDuration("listen.gso_flush_timeout", defaultGSOFlushTimeout)
+	if timeout < 0 {
+		timeout = 0
+	}
+	u.gsoFlushTimeout = timeout
+}
+
+func (u *StdConn) setGROBufferSize(size int) {
+	if size < MTU {
+		size = defaultGROReadBufferSize
+	}
+	if size > maxGSOBatchBytes {
+		size = maxGSOBatchBytes
+	}
+	u.groBufSize = size
+}
+
+func (u *StdConn) disableGSO() {
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+	u.enableGSO = false
+	_ = u.flushGSOlocked()
+	u.gsoBuf = nil
+	u.gsoSegments = 0
+	u.gsoSegSize = 0
+	u.stopGSOTimerLocked()
 }
 
 func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {
@@ -305,7 +542,239 @@ func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {
 	return nil
 }
 
+func (u *StdConn) queueGSOPacket(b []byte, addr netip.AddrPort) error {
+	if len(b) == 0 {
+		return nil
+	}
+
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+
+	if !u.enableGSO || !addr.IsValid() || len(b) > u.gsoMaxBytes {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		return errGSOFallback
+	}
+
+	if u.gsoSegments == 0 {
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	} else if addr != u.gsoAddr || len(b) != u.gsoSegSize {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	}
+
+	if len(u.gsoBuf)+len(b) > u.gsoMaxBytes {
+		if err := u.flushGSOlocked(); err != nil {
+			return err
+		}
+		if cap(u.gsoBuf) < u.gsoMaxBytes {
+			u.gsoBuf = make([]byte, 0, u.gsoMaxBytes)
+		}
+		u.gsoAddr = addr
+		u.gsoSegSize = len(b)
+	}
+
+	u.gsoBuf = append(u.gsoBuf, b...)
+	u.gsoSegments++
+
+	if u.gsoSegments >= u.gsoMaxSegments || u.gsoFlushTimeout <= 0 {
+		return u.flushGSOlocked()
+	}
+
+	u.scheduleGSOFlushLocked()
+	return nil
+}
+
+func (u *StdConn) flushGSOlocked() error {
+	if u.gsoSegments == 0 {
+		u.stopGSOTimerLocked()
+		return nil
+	}
+
+	payload := append([]byte(nil), u.gsoBuf...)
+	addr := u.gsoAddr
+	segSize := u.gsoSegSize
+
+	u.gsoBuf = u.gsoBuf[:0]
+	u.gsoSegments = 0
+	u.gsoSegSize = 0
+	u.stopGSOTimerLocked()
+
+	if segSize <= 0 {
+		return errGSOFallback
+	}
+
+	err := u.sendSegmented(payload, addr, segSize)
+	if errors.Is(err, errGSODisabled) {
+		u.l.WithField("addr", addr).Warn("UDP GSO disabled by kernel, falling back to sendto")
+		u.enableGSO = false
+		return u.sendSegmentsIndividually(payload, addr, segSize)
+	}
+
+	return err
+}
+
+func (u *StdConn) sendSegmented(payload []byte, addr netip.AddrPort, segSize int) error {
+	if len(payload) == 0 {
+		return nil
+	}
+
+	control := make([]byte, unix.CmsgSpace(2))
+	hdr := (*unix.Cmsghdr)(unsafe.Pointer(&control[0]))
+	hdr.Level = unix.SOL_UDP
+	hdr.Type = unix.UDP_SEGMENT
+	setCmsgLen(hdr, unix.CmsgLen(2))
+	binary.NativeEndian.PutUint16(control[unix.CmsgLen(0):unix.CmsgLen(0)+2], uint16(segSize))
+
+	var sa unix.Sockaddr
+	if addr.Addr().Is4() {
+		var sa4 unix.SockaddrInet4
+		sa4.Port = int(addr.Port())
+		sa4.Addr = addr.Addr().As4()
+		sa = &sa4
+	} else {
+		var sa6 unix.SockaddrInet6
+		sa6.Port = int(addr.Port())
+		sa6.Addr = addr.Addr().As16()
+		sa = &sa6
+	}
+
+	if _, err := unix.SendmsgN(u.sysFd, payload, control, sa, 0); err != nil {
+		if errno, ok := err.(syscall.Errno); ok && (errno == unix.EINVAL || errno == unix.ENOTSUP || errno == unix.EOPNOTSUPP) {
+			return errGSODisabled
+		}
+		return &net.OpError{Op: "sendmsg", Err: err}
+	}
+	return nil
+}
+
+func (u *StdConn) sendSegmentsIndividually(buf []byte, addr netip.AddrPort, segSize int) error {
+	if segSize <= 0 {
+		return errGSOFallback
+	}
+
+	for offset := 0; offset < len(buf); offset += segSize {
+		end := offset + segSize
+		if end > len(buf) {
+			end = len(buf)
+		}
+		var err error
+		if u.isV4 {
+			err = u.writeTo4(buf[offset:end], addr)
+		} else {
+			err = u.writeTo6(buf[offset:end], addr)
+		}
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (u *StdConn) scheduleGSOFlushLocked() {
+	if u.gsoTimer == nil {
+		u.gsoTimer = time.AfterFunc(u.gsoFlushTimeout, u.gsoFlushTimer)
+		return
+	}
+	u.gsoTimer.Reset(u.gsoFlushTimeout)
+}
+
+func (u *StdConn) stopGSOTimerLocked() {
+	if u.gsoTimer != nil {
+		u.gsoTimer.Stop()
+		u.gsoTimer = nil
+	}
+}
+
+func (u *StdConn) gsoFlushTimer() {
+	u.gsoMu.Lock()
+	defer u.gsoMu.Unlock()
+	_ = u.flushGSOlocked()
+}
+
+func parseGROControl(control []byte) (int, int) {
+	if len(control) == 0 {
+		return 0, 0
+	}
+
+	cmsgs, err := unix.ParseSocketControlMessage(control)
+	if err != nil {
+		return 0, 0
+	}
+
+	for _, c := range cmsgs {
+		if c.Header.Level == unix.SOL_UDP && c.Header.Type == unix.UDP_GRO && len(c.Data) >= 2 {
+			segSize := int(binary.NativeEndian.Uint16(c.Data[:2]))
+			segCount := 0
+			if len(c.Data) >= 4 {
+				segCount = int(binary.NativeEndian.Uint16(c.Data[2:4]))
+			}
+			return segSize, segCount
+		}
+	}
+
+	return 0, 0
+}
+
+func (u *StdConn) emitGROSegments(r EncReader, addr netip.AddrPort, payload []byte, segSize int) bool {
+	if segSize <= 0 {
+		return false
+	}
+
+	for offset := 0; offset < len(payload); offset += segSize {
+		end := offset + segSize
+		if end > len(payload) {
+			end = len(payload)
+		}
+		segment := make([]byte, end-offset)
+		copy(segment, payload[offset:end])
+		r(addr, segment)
+	}
+	return true
+}
+
+func normalizeGROSegSize(segSize, segCount, total int) int {
+	if segSize <= 0 || total <= 0 {
+		return segSize
+	}
+
+	if segSize > total && segCount > 0 {
+		segSize = total / segCount
+		if segSize == 0 {
+			segSize = total
+		}
+	}
+
+	if segCount <= 1 && segSize > 0 && total > segSize {
+		calculated := total / segSize
+		if calculated <= 1 {
+			calculated = (total + segSize - 1) / segSize
+		}
+		if calculated > 1 {
+			segCount = calculated
+		}
+	}
+
+	if segSize > MTU {
+		return MTU
+	}
+
+	return segSize
+}
+
 func (u *StdConn) Close() error {
+	u.disableGSO()
 	return syscall.Close(u.sysFd)
 }
 

udp/udp_linux_32.go

@@ -30,13 +30,16 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int, bufSize int) ([]rawMessage, [][]byte, [][]byte) {
+	if bufSize <= 0 {
+		bufSize = MTU
+	}
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		buffers[i] = make([]byte, bufSize)
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
@@ -52,3 +55,25 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 
 	return msgs, buffers, names
 }
+
+func setRawMessageControl(msg *rawMessage, buf []byte) {
+	if len(buf) == 0 {
+		msg.Hdr.Control = nil
+		msg.Hdr.Controllen = 0
+		return
+	}
+	msg.Hdr.Control = &buf[0]
+	msg.Hdr.Controllen = uint32(len(buf))
+}
+
+func getRawMessageControlLen(msg *rawMessage) int {
+	return int(msg.Hdr.Controllen)
+}
+
+func getRawMessageFlags(msg *rawMessage) int {
+	return int(msg.Hdr.Flags)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, l int) {
+	h.Len = uint32(l)
+}

udp/udp_linux_64.go

@@ -33,13 +33,16 @@ type rawMessage struct {
 	Pad0 [4]byte
 }
 
-func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
+func (u *StdConn) PrepareRawMessages(n int, bufSize int) ([]rawMessage, [][]byte, [][]byte) {
+	if bufSize <= 0 {
+		bufSize = MTU
+	}
 	msgs := make([]rawMessage, n)
 	buffers := make([][]byte, n)
 	names := make([][]byte, n)
 
 	for i := range msgs {
-		buffers[i] = make([]byte, MTU)
+		buffers[i] = make([]byte, bufSize)
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
@@ -55,3 +58,25 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 
 	return msgs, buffers, names
 }
+
+func setRawMessageControl(msg *rawMessage, buf []byte) {
+	if len(buf) == 0 {
+		msg.Hdr.Control = nil
+		msg.Hdr.Controllen = 0
+		return
+	}
+	msg.Hdr.Control = &buf[0]
+	msg.Hdr.Controllen = uint64(len(buf))
+}
+
+func getRawMessageControlLen(msg *rawMessage) int {
+	return int(msg.Hdr.Controllen)
+}
+
+func getRawMessageFlags(msg *rawMessage) int {
+	return int(msg.Hdr.Flags)
+}
+
+func setCmsgLen(h *unix.Cmsghdr, l int) {
+	h.Len = uint64(l)
+}

udp/udp_rio_windows.go

@@ -134,7 +134,7 @@ func (u *RIOConn) bind(sa windows.Sockaddr) error {
 	return nil
 }
 
-func (u *RIOConn) ListenOut(r EncReader) {
+func (u *RIOConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	for {